Scientific Opinion on Evaluation of the Toxicological Relevance of Pesticide Metabolites for Dietary Risk Assessment 1

Transcription

1 EFSA Journal 2012;10(07):2799 SCIENTIFIC PININ Scientific pinion on Evaluation of the Toxicological Relevance of Pesticide Metabolites for Dietary Risk Assessment 1 EFSA Panel on Plant Protection Products and their Residues (PPR) 2, 3 European Food Safety Authority (EFSA), Parma, Italy This scientific output, published on 02/08/2012, replaces the earlier version published on 26/07/2012. ABSTRACT The European Food Safety Authority (EFSA) asked the Panel on Plant Protection Products and their Residues (PPR) to develop an opinion on approaches to evaluate the toxicological relevance of metabolites and degradates of pesticide active substances in dietary risk assessment. This opinion identifies the threshold of toxicological concern (TTC) concept as an appropriate screening tool. The TTC values for genotoxic and toxic compounds were found to be sufficiently conservative for chronic exposure, as a result of a validation study with a group of pesticides belonging to different chemical classes. Three critical steps were identified in the application of a TTC scheme: 1) the estimate of the level of the metabolite, 2) the evaluation of genotoxicity alerts and 3) the detection of neurotoxic metabolites. Tentative TTC values for acute exposure were established by the PPR Panel by analysis of the lowest 5 th percentiles of No bserved Adverse Effect Levels (NAELs) used to establish the Acute Reference Doses (ARfD) for the EFSA pesticide data set. Assessment schemes for chronic and acute dietary risk assessment of pesticide metabolites, using the TTC approach and combined (Q)SAR and read across, are proposed. The opinion also proposes how the risk assessment of pesticide metabolites that are stereoisomers should be addressed due to isomer ratio changes reflected in the composition of metabolites. The approach is ready for use, but it is anticipated that on many occasions the outcome of the assessment scheme will be that further testing is needed to reach a firm conclusion 1 n request from EFSA, Question No EFSA-Q , adopted on 21 June Panel members: Jos Boesten, Claudia Bolognesi, Theo Brock, Ettore Capri, Anthony Hardy, Andrew Hart, Karen Ildico Hirsch-Ernst, Susanne Hougaard Bennekou, Michael Klein, Robert Luttik, Kyriaki Machera, Angelo Moretto (until January 2011), Bernadette ssendorp, Annette Petersen, Yolanda Pico, Andreas Schäffer, Paulo Sousa, Walter Steurbaut, Anita Strömberg, Maria Tasheva, Ton van der Linden, Christiane Vleminckx. ne member of the Panel did not participate in parts of the discussion on the subject referred to above because of potential conflicts of interest identified in accordance with the EFSA policy on declarations of interests. Correspondence: pesticides.ppr@efsa.europa.eu 3 Acknowledgement: EFSA wishes to thank the members of the Working Group on Toxicological Relevance of Pesticide Metabolites: Susan Barlow, Emilio Benfenati, Claudia Bolognesi, Alan Boobis, Karen Ildico Hirsch-Ernst, Susanne Hougaard Bennekou, Kyriaki Machera, tto Meyer, Angelo Moretto (until January 2011), Andre Muller, Markus Müller, Bernadette ssendorp, Yolanda Pico, Rebecca Scrivens, Anita Strömberg, Maria Tasheva and Christiane Vleminckx, the hearing experts Albert Bergmann, Ian Dewhurst and Andrew Worth and the EFSA staff Lazlo Bura, Edgars Felkers (until March 2012), Miriam Jacobs, István Sebestyén (until May 2012), Hans Steinkellner and Manuela Tiramani for the support provided to this scientific opinion. Suggested citation: EFSA Panel on Plant Protection Products and their Residues (PPR); Scientific pinion on Evaluation of the Toxicological Relevance of Pesticide Metabolites for Dietary Risk Assessment. EFSA Journal 2012;10(07): [187 pp.] doi: /j.efsa Available online: European Food Safety Authority, 2012

2 on the toxicological relevance of the metabolite. However, the benefit of applying the approach is that it will allow prioritisation of metabolites for subsequent testing. EFSA will develop a Guidance Document based on the results in this opinion. European Food Safety Authority, 2012 KEY WRDS Pesticide metabolites, dietary exposure, endocrine disruptor, Quantitative Structure Activity Relationship, (Q)SAR, read-across, stereoisomer, Threshold of Toxicological Concern (TTC) EFSA Journal 2012;10(07):2799 2

3 SUMMARY The use of pesticides in agriculture may lead to a large number of metabolites being present at low levels in food and feed. Progress in analytical methods and their increasing sensitivity results in the detection of a growing number of metabolites in low amounts. The residue definition for dietary risk assessment should include the active substance and all metabolites of toxicological relevance. A comprehensive toxicological dossier is developed for parent compounds, prior to approval of substances for use within EU (Regulation EC (No) 1107/2009), while often only limited information about the toxicological properties of metabolites is available. In light of these considerations, EFSA asked the PPR Panel to develop an opinion on approaches to evaluate the toxicological relevance of metabolites and degradates of pesticide active substances in dietary risk assessment. The need to minimise the use of laboratory animals where possible was highlighted. The Panel was also asked to consider whether the approaches and methodologies developed for pesticide metabolites are applicable to isomer ratio changes of active substances existing as isomer mixtures or which are used as individual isomers. A key issue is whether a metabolite would have been tested with the parent compound in laboratory species, due to its formation in vivo. For those metabolites not so tested, because they are unique to plants or livestock, an alternative approach is necessary. The PPR Panel considered relevant publications in the scientific literature and current applications of non-testing approaches in various regulatory contexts. n this basis four projects were outsourced to evaluate the potential impact of metabolic processes on the toxicity of pesticide metabolites and to explore the reliability of the available computational tools. The PPR Panel has developed a strategy to estimate the dietary exposure to pesticide metabolites. Several exposure scenarios were considered covering various possibilities of metabolite ratio extrapolation and the extent of uses. Case studies illustrate the methods that have been used. It is noted that the choice of the scenario has a considerable impact on the estimate of the metabolite level to be used. The Panels conclusions on these approaches are as follows: The potential impact of structural metabolic changes to parent compounds on the toxicological properties of derived metabolites was analysed for the most relevant chemical classes of active substances listed in Annex 1 of Directive 91/414/EEC. Despite the high level of uncertainty due to the heterogeneity of ADME studies and inadequacy of toxicological data on metabolites, the metabolic pathways are in most cases specific for each chemical group and toxification/detoxification potential cannot be reliably attributed to specific metabolic steps. The TTC concept is the most appropriate tool for evaluating the toxicological relevance of pesticide metabolites. The existing TTC values for genotoxic and toxic compounds were found to be sufficiently conservative for chronic exposure by a validation study with groups of pesticides belonging to different chemical classes. These values, based on the assumption of continuous exposure during lifetime, are overly conservative for short term exposure duration. Tentative TTC values for acute exposure were established by the analysis of the lowest 5 th percentiles of NAELs used to establish ARfDs for the EFSA pesticide data set. Three critical steps were identified in the application of the TTC scheme in risk assessment of pesticide metabolites: 1) the estimate of the level of the metabolite, 2) the evaluation of genotoxicity alerts and the 3) detection of neurotoxic metabolites arising from a parent compound with a structural alert not covered by the scheme. The evaluation of genotoxicity alerts was addressed in an outsourced project involving the application of several (Q)SAR models using the largest dataset available of active ingredients and metabolites. The results showed individual models to have low sensitivities in identifying genotoxic pesticides, while the same tools applied in combination appeared good identifiers of classified mutagens. The low sensitivity was mainly attributed to the heterogeneity of the EFSA Journal 2012;10(07):2799 3

4 underlying pesticide database. The PPR Panel concluded that the performance of these applied tools is not satisfactory and cannot support, at the present time, the application of solely (Q)SAR approaches to predict the potential genotoxicity of unknown pesticide metabolites. The applicability of (Q)SAR tools, grouping and read-across approaches in the evaluation of and neurotoxic effects of pesticide metabolites was addressed by another outsourced project. The predictivity for neurotoxicity of the (Q)SAR models, tested alone or in combination, is currently inadequate to be applied for pesticide metabolites. (Q)SAR tools alone appeared insufficiently reliable to predict effects, due to their low sensitivity and specificity, but a stepwise approach involving (Q)SAR analysis and readacross, resulted in an improvement in the identification of potential toxicants. The results of the (Q)SAR projects allowed the PPR Panel to propose the application of computational methods, involving the separate or sequential use of (Q)SAR and read-across in the prediction of genotoxicity and toxicity, to complement the TTC approach in the assessment scheme for pesticide metabolite exposure. Estimates of exposure to pesticide metabolites by the Panel are based mainly on residue metabolism studies. These data have also been adapted using a metabolite to parent ratio applied to the available residue end-points from the supervised trials data to give different estimates of exposure for both chronic and acute exposure. The key issue affecting the results is the potential for extrapolating data, encompassing metabolism groupings and the extent of uses. The approaches tested allowed the Panel to propose a dietary exposure tree for pesticide metabolites. However different methodological approaches produce different outcomes and risk managers would need to advise on the level of protection that is desired. The scientific principles that underpin pesticide metabolite exposure calculations (above) are also directly relevant to the derivation of conversion factors which are established during the regulatory evaluation of parent compounds in the framework of Regulation (EC) No 1107/2009 when the residue definitions for monitoring and dietary risk assessment differ. The PPR Panel recognises that currently, there is no unambiguous approach to deriving conversion factors and recommends the developing further guidance in this area. Chronic and acute assessment schemes are proposed for the risk assessment of pesticide metabolites considering different strategies for mammalian (rodent or laboratory test species) and plant or livestock specific metabolites. A chronic exposure estimate is necessary in all cases, while an acute exposure assessment is needed only when an Acute Reference Dose (ARfD) has been allocated for the parent compound or structural alerts for acute neurotoxicity and toxicity are detected. The chronic assessment scheme involves the comparison of chronic exposure with the corresponding threshold values given in the decision tree. Computational tools involving the combination of (Q)SAR and read-across are proposed in the evaluation of an alert for genotoxicity. If the exposure estimate exceeds the identified TTC values, different approaches are proposed for mammalian rodent and plant or livestock metabolites. A weight of evidence approach is recommended to determine if the toxicological profile of rodent metabolites is covered by the data on parent compound. Plant or livestock specific metabolites need to be assessed using an appropriate testing strategy. An acute exposure assessment scheme was developed by the PPR Panel. Ad hoc acute TTC values of 0.3 µg/kg bw/d for substances with a neurotoxicity alert and 5 µg/kg bw/d for substances allocated in Cramer class II and III were derived. A combination of (Q)SAR and read-across approaches is proposed for the prediction of toxicity. EFSA Journal 2012;10(07):2799 4

5 Where exposure to a metabolite exceeds the respective TTC value, acute and chronic toxicity testing strategies were proposed by the PPR Panel, considering the need to derive health based limits for human exposure. The opinion also proposes how the risk assessment of pesticide metabolites that are stereoisomers should be addressed due to isomer ratio changes reflected in the composition of metabolites. The PPR Panel does not propose that the TTC scheme is used for individual stereoiosomers, although the TTC scheme has utility as a screening assessment of the isomer mixture for a metabolite. Further development of (Q)SAR tools would be beneficial, both to predict genotoxicity and to address stereochemistry aspects. Furthermore, metabolism guidelines should require compositional information on stereochemistry to consider the full impact on the dietary risk assessment. The approaches described in this opinion are ready for use, but it is anticipated that on many occasions the outcome of the assessment scheme will be that further testing is needed to reach a firm conclusion on the toxicological relevance of the pesticide metabolite. However, the benefit of applying the approaches is that it will allow prioritisation of pesticide metabolites for subsequent testing. These approaches should not be used as an alternative to conventional risk assessment for pesticide active substances (parent compounds) themselves occurring as residues in food. They should be assessed prior to authorisation on the basis of dossiers including toxicological tests (Regulation (EC) No 1107/2009). It is noted that EFSA will develop a Guidance Document based on the results in this opinion. EFSA Journal 2012;10(07):2799 5

6 TABLE F CNTENTS Abstract... 1 Summary... 3 Table of contents... 6 Background as provided by EFSA... 9 Terms of reference as provided by EFSA Assessment Introduction Structure of the pinion Current approaches to residue definition for pesticides Use of terms pesticide residue and residue definition Residue definitions Comparison of residue definitions by EFSA and JMPR Current status and future development of analytical methods for establishing residue definitions Relevant metabolites Rationale for the outsourced projects Impact of metabolic processes on the toxicological properties of pesticide residues Evaluation of toxicological profiles of pesticide metabolites in different regulatory contexts (results of outsourced project) Evaluation of metabolic pathways with respect to their toxification/detoxification potential for selected groups of pesticides as presented by the contractor Conclusions of the PPR Panel Evaluation of metabolic pathways with respect to their toxification/detoxification potential for selected groups of pesticides Considerations by the PPR Panel on conjugated and bound residues Evaluation of toxicological profiles of pesticide metabolites Threshold of Toxicological Concern (TTC) Current applications of TTC concept Additives: JECFA and EFSA Pharmaceuticals: European Medicines Agency (EMA) Pesticide metabolites in groundwater Guidance document SANC/221/2000 rev 10 of 25 February Industrial chemicals (REACH): ECHA Applicability of TTC approach in the dietary risk assessment of pesticide metabolites: outcome of the outsourced project Validation of the developed TTC concept: case studies on metabolites Estimation of the exposure Results and conclusions of TTC case study presented by the contractor Conclusions of the PPR Panel on the TTC case study presented by the contractor Application of TTC approach for acute exposure Derivation of acute exposure thresholds verall conclusions on TTC SAR/ (Q)SAR concept Characterisation of chemical space Performance of (Q)SAR models Read-across (Q)SAR approach in the dietary risk assessment of pesticide metabolites Software tools for genotoxicity and carcinogenicity prediction Current applications of (Q)SAR approach in predicting mutagenicity and carcinogenicity Applicability of (Q)SAR analysis to the evaluation of the toxicological relevance of metabolites of pesticide active substances for dietary risk assessment utsourced project by the Joint Research Centre (JRC) Framework for assessing the usefulness of (Q)SAR models EFSA Journal 2012;10(07):2799 6

7 6.6. Applicability of (Q)SAR analysis to the evaluation of genotoxicity of pesticide metabolites for dietary risk assessment: case studies Software tools applied Compilation of datasets Characterization of chemical space by Principal Component Analysis (PCA) Performance of the models in predicting mutagenicity Prediction results Conclusions of the contractor Conclusion by PPR Panel Applicability of (Q)SAR analysis in the evaluation of and neurotoxicity effects of pesticide metabolites: outcomes of outsourced activity Predictive performance of (Q)SAR/read-across strategy for neurotoxicity Predictive performance of (Q)SAR/read-across strategy for toxicity Conclusions of the contractor Conclusions by the PPR Panel Potential exposure to pesticide metabolites in the human diet Introduction Types of exposure scenarios Estimation of metabolite levels Models used for calculating acute and chronic dietary exposure Conversion factors for estimating metabolite levels Results and discussion Impact of stereochemistry on the toxicological relevance of pesticide metabolites for dietary risk assessment Terminology Introduction Consequences of stereoisomerism for risk assessment Data requirements regarding stereochemistry under Regulation (EC) 1107/ Assessment of the toxicological relevance of isomer ratio changes of metabolites Exposure assessment Conclusion on applicability of approaches to address relevance of metabolites to isomer ratio changes Critical issues and uncertainties Critical issues in toxicology TTC for genotoxicity TTC for neurotoxicity TTC for endocrine disruptors New TTC values for acute exposure Critical issues on exposure Ratio metabolite/parent estimations for estimating the exposure of pesticide metabolites Extent of uses of the pesticide General recommendations on these two issues: Use of residue data for metabolite estimation Acute exposure to metabolites Cumulative and aggregate exposure Critical issues on stereoisomers Uncertainties affecting the assessment Proposed strategy for assessing the toxicological relevance of pesticide metabolites Assessment scheme Assessment scheme for chronic exposure Assessment scheme for acute exposure Acute and chronic toxicity - Testing Strategy Genotoxicity testing Toxicity testing CNCLUSINS AND RECMMENDATINS EFSA Journal 2012;10(07):2799 7

8 References Appendices A. European Guidelines in international context B. Selected Residue definition examples of pesticides with supporting analytical methods for residue monitoring/enforcement C. EFSA and JMPR residue definitions on pesticides for consumer risk assessment D. Examples of metabolite to parent ratio for 10 pesticides and its dependence on use pattern E. Pesticide metabolite estimations - Case studies F. Checklist in support of assessment of adequacy of (Q)SAR Predictions G. Acute exposure thresholds Glossary Abbreviations EFSA Journal 2012;10(07):2799 8

9 BACKGRUND AS PRVIDED BY EFSA Annex VI of Council Directive 91/414/EEC of 15 July 1991 concerning the placing on the market of plant protection products 45 sets out uniform principles for the evaluation and authorisation of chemical plant protection products and the active substances they contain. The likely risk to humans, animals and the environment need to be addressed. Assessment of the risk for the consumer is a major part of this process. This assessment requires the identification of metabolites and of degradates of the active substances present in food commodities. Metabolites may be produced from plant metabolism in primary and following crops, from microbiological activity in soil, or from livestock metabolism after consumption of feeding stuffs containing residues. Degradates arise from abiotic physical and chemical processes (e.g. photolysis) and from processing before the consumption of plant and animal commodities (e.g. cooking). In practice the consumer is therefore exposed not only to the active substance as applied, but also to a wide range of chemical compounds as a result of metabolic and degradation processes. The number and amount of distinct compounds, defining the residue pattern the consumer is exposed to, may widely differ from pesticide to pesticide depending on many parameters. ne of the outcomes of the evaluation of an application for use of an active substance on a crop is the establishment of two residue definitions, one for monitoring and one for dietary risk assessment. As outlined in the guidance document6 on residue definition, adopted by ECD in 2006, the underlying rationales for these two definitions are different. While the residue definition for monitoring has regulatory purposes for the enforcement of the MRLs (Maximum Residue Levels) and must reflect analytical practicalities, the residue definition for dietary risk assessment may be wider, as its purpose is to assess consumer safety, and it should therefore include all metabolites and degradates of toxicological relevance. In other words, in order to perform an appropriate assessment of the risk for the consumer, the residue definition for dietary risk assessment should be qualitatively and quantitatively representative of the actual toxicological burden. This means that establishment of the residue definition for dietary risk assessment requires not only a decision on which metabolites or degradates, due to their level, may significantly contribute to toxicological effects, but also an assessment of the toxicological endpoints of interest, and related reference values. A major difficulty stems from the fact that, from the mixture (active substance, its metabolites and degradates) to which the consumer is exposed, only the toxicological properties of the active substance are in practice directly investigated through the range of toxicological studies required by Directive 91/414/EEC. In contrast, very limited information about the toxicological properties of metabolites and degradates is available in the majority of cases, while requests for further toxicological studies are restricted as far as possible to minimise the use of animals in toxicological testing. In view of the guidance document on residue definition published by the ECD, and in order to ensure consistency and robustness of expert judgement, EFSA considers that all relevant scientific tools need to be reviewed and evaluated so that they can be used optimally in evaluating the toxicological burden of metabolites and degradates. Following adoption of this opinion, a guidance document will be developed on the establishment of residue definition for dietary risk assessment. This guidance should be a practical instrument, aimed at helping risk assessors and regulatory authorities to adopt such definitions based on a combination of 4 EC (1991). Council Directive 91/414/EEC of 15 July 1991 concerning the placing of plant protection products on the market. fficial Journal L 230, August In the course of drafting this opinion, this Directive was replaced by Regulation (EC) No 1107/ ECD, Environmental Directorate (Guidance document ont the definition of residue, Series on testing and assessment Nr 63. Series on Pesticides Nr. 31, 10-ct EFSA Journal 2012;10(07):2799 9

10 scientific tools. This guidance should also be used for identifying cases where further experimental data are needed. TERMS F REFERENCE AS PRVIDED BY EFSA EFSA asked the PPR Panel to develop an opinion on approaches to evaluating the toxicological relevance of metabolites and degradates of pesticide active substances in dietary risk assessment. The original terms of reference were extended to address also the issue of isomer conversion and therefore amended as follows: Regarding possible isomer ratio changes of the active substances existing as isomer mixtures or which is an individual isomer, the PPR Panel is asked to address: If the approaches and methodologies developed for pesticide metabolites are applicable to the isomer ratio changes of the active substances existing as isomer mixtures or which are an individual isomer To identify if relevant, specific issues of dietary risk assessment applicable to active substances existing as isomer mixtures or which are an individual isomer and develop the respective appropriate assessment methodologies or identification of relevant research need. ASSESSMENT 1. Introduction The use of pesticides on food and feed crops may lead to residues in edible parts of the plant and hence results in exposure of the consumer to a mixture of compounds including the active substance and/or its metabolite(s), (ECD, 2009a). The number of metabolites varies from pesticide to pesticide and from none to, in some cases, a large array of metabolites found. In addition progress in analytical methods and their increasing sensitivity results in the detection of a growing number of metabolites at low levels. The term metabolite in this opinion refers to a metabolite or a degradation product of an active substance as defined in Regulation (EC) No 1107/ (see Glossary). Metabolites have varying relevance for human exposure depending on their inherent toxicities and levels at which they are found. The process of metabolism or degradation of active compounds may give breakdown products maintaining the active moiety responsible for the biological activity and in some cases for the toxic effects, or alternatively the toxic moiety may be modified to reduce or eliminate toxicity. Also a new toxic moiety may be created with a potentially different mechanism of action. Therefore metabolism studies in soil, plants, and livestock using radiolabelled active substances are requested prior to the authorisation of plant protection products and the active substance that they contain. The objective of these studies is to identify the nature of terminal residues in food and feed commodities (from plant and animal origin) and quantify them. Depending on a number of factors (e.g. mode and time of application, environmental conditions, nature of the crop), the terminal residues may differ between crops, and between crops and animal products. In addition, the residue pattern in rotational crops frequently differs from that in primary crops, which is often related to the residues in the soil after an aging period and uptake by plants. A crop metabolism study should be submitted for each crop group for which use is proposed. Similarly, depending on potential exposure of livestock, metabolism studies in livestock, e.g. lactating 7 Regulation (EC) No 1107/2009 of the European Parliament and of the Council of 21 ctober 2009 concerning the placing of plant protection products on the market and repealing Council Directives 79/117/EC and 91/414/EEC. fficial Journal L 309, November EFSA Journal 2012;10(07):

11 goat and laying hen, should be provided. The independent assessment of individual plant and livestock metabolism studies in conjunction with proposed analytical methods for enforcement of MRLs may lead to different conclusions regarding the residue definition for monitoring and for risk assessment in individual plant and animal commodities. Nevertheless, for reasons of pragmatism it is current practice to try to establish common residue definitions for monitoring and for risk assessment (unless suitable conversion factors can be proposed), covering plant and animal products. As the use pattern of pesticides is an evolving process, residue definitions may need to be re-evaluated periodically with the development of new uses for national registration. These re-evaluations affect not only raw plant commodities, but also processed commodities and products of animal origin which may result in a change in the number of metabolites in the residue definitions. Similarly, the evaluation of compounds under Directive 91/414/EEC (and now under Regulation EC (No) 1107/2009) relying on the one representative use concept for approval in the EU. Article 4 paragraph 5 of Regulation in the regulation may lead to conclusions based on the evaluation of only a limited number of the existing range of uses of active substances. When additional uses are assessed, novel metabolites may be identified which affect the residue definition. A comprehensive toxicological dossier (data requirements according to Regulation (EC) No 544/ ) must be developed for parent compounds prior to approval of substances for use within the EU, including toxicokinetic and metabolism studies in mammals. However, specific toxicity studies may be available on only some metabolites, as these data tend not to be provided for the full range of metabolites found. The extent of testing necessary will depend on whether the metabolite is produced at appreciable levels in laboratory species, where the toxicological effects will reflect, at least in part, the toxicity of the metabolite. In contrast, where a metabolite is unique to plants or livestock, no information on potential toxicity will be available from the toxicity testing of the parent compound. Considering the limited toxicity testing resources worldwide and in order to minimise the use of laboratory animals in toxicological testing, new approaches should be considered for the risk assessment of metabolites, taking into account all the available information and using predictive models based on comparative analyses of hazard data from structurally related compounds. In this context, all available alternative scientific tools, need to be reviewed and evaluated for their applicability in the evaluation of the toxicological profile of metabolites of pesticides, so that they can be optimally applied to derive relevant toxicological (threshold) values. A new Guidance document on the definition of pesticide residues for monitoring and risk assessment was adopted by ECD in 2006, with a slightly revised version published in The most recent Guidance documents available on the definition of residues (e.g. FA, 2002; EC, 1997a, b) were taken into account by the ECD during the drafting of this guidance and the FA manual dated 2002 was the main document from which the ECD Guidance was developed. The ECD Guidance recommends that the residue definition for consumer risk assessment should include those metabolites which, due to their levels present, significantly contribute to the dietary risk. However it does not present tools to evaluate the toxicological burden of pesticide metabolites. Marketed pesticides can comprise various types of stereoisomer composition: a single mixture, various different mixtures, or a single isomer. Each case should be handled differently. With respect to the assessment of isomer mixtures, the ECD Guidance states that in practice the starting point in authorising plant protection products is the mixture of isomers where all metabolites should be found and taken into account. This can be translated into: the composition of the mixture of stereoisomers in a technical active ingredient has to be known and linked to the hazard and risk profile of the pesticide under consideration. However, the ECD Guidance identifies several aspects that should be considered in deciding whether isomers need special consideration: The type of isomers (enantiomers, diastereomers or cis-trans isomers) should be clarified. Stability of the isomers (inter-conversion). Level of isomers 8 Regulation (EU) No 544/2011 of 10 June 2011 implementing Regulation (EC) No 1107/2009 of the European Parliament and of the Council as regards the data requirements for active substances. fficial Journal L 155, June EFSA Journal 2012;10(07):

12 Differences in their toxicological properties 1.1. Structure of the pinion In the current chapter and in chapter 2 supported by appendices A-C the scope of the opinion is presented as well as background and current approaches on how definitions of pesticide residues are derived. Chapter 3 describes the rationales for four outsourced projects on different tools not involving animal testing to evaluate the potential toxicological impact of metabolites from pesticides. The toxicological consequence of metabolism of the active substance is presented and discussed in chapter 4. The Threshold of Toxicological Concern (TTC) concept and a proposal for a modified TTC approach (including both chronic and acute exposure thresholds) for application to pesticide metabolites is presented in chapter 5 and Appendix G. The application of Quantitative Structure Activity Relationship (Q)SAR) methods, with particular reference to genotoxicity alerts, is discussed in chapter 6 and Appendix F. The applicability of (Q)SAR and read-across methods for evaluating and neurotoxic effects of metabolites is presented in chapter 7. To perform a risk assessment, estimates of consumer intakes of all possible pesticide metabolites are required. Therefore, in chapter 8, supported by Appendices D-E metabolite exposure predictions of laboratory animal metabolites and metabolites that are specific to plants and livestock are presented. Conversion factors for converting residues determined in the residue definition for monitoring to values suitable for a dietary risk assessment are also discussed in this chapter. In chapter 9 the applicability of the approaches and methodologies (as developed for pesticide metabolites) to isomer mixtures is discussed. Critical issues as well as uncertainties related to the different chapters in the opinion are presented in chapter 10. In chapter 11 a strategy for assessing the toxicological relevance of pesticide metabolites is proposed. Finally chapter 12 gives conclusions and recommendations for future approaches and research. 2. Current approaches to residue definition for pesticides 2.1. Use of terms pesticide residue and residue definition Many Guidelines, Guidance documents and Regulations covering pesticide residues are available (ECD, 2009a; FA, 2009, 2012; EC, 1997a, b; Regulation (EC) No 1107/ ), Regulation (EC) No 396/ ), that differ to some extent from each other on use of terms describing residues. This opinion relates to Regulation 1107/2009, which stipulates that substances or products produced or placed on the market should not have any harmful effect on human or animal health or any unacceptable effects on the environment. The term pesticide residue in Regulation EC (No) 1107/2009 is defined as one or more substances present in or on plants and plant products, edible animal products, drinking water or elsewhere in the environment and resulting from the use of a plant protection product, including their metabolites, breakdown or reactions products 11. Most active substances undergo, after application of the formulated product to crops, chemical and biochemical degradation processes generally leading to an overall reduction in residue levels. At the same time (relevant) metabolites and degradates may be formed. The "residue definition" aims to provide a reasonable description of compounds related to the active ingredient initially applied and contributing to toxicological burden when the food items are consumed --> residue definition for risk assessment 9 Regulation (EC) No 1107/2009 of the European Parliament and of the Council of 21 ctober 2009 concerning the placing of plant protection products on the market and repealing Council Directives 79/117/EC and 91/414/EEC. fficial Journal L 309, November Regulation (EC) No 396/2005 of the European Parliament and of the Council of 23 February 2005 on maximum residue levels of pesticides in or on food and feed of plant and animal origin and amending Council Directive 91/414/EEC. fficial Journal L70, March Breakdown or reaction products is in this opinion are referred to as degradates. EFSA Journal 2012;10(07):

13 and/or being applicable to routine residue analytical methodology --> residue definition for monitoring/enforcement of residues present in the food item at time of harvest or slaughter The analytical methodology used in developing the residue data for submission of a dossier is usually more complex and demanding than that for routine monitoring (ECD, 2009b) and is more likely to provide direct information on metabolite levels. Bridging between the residue definition for monitoring/enforcement and the residue definition for risk assessment is achieved by applying a socalled conversion factor (see chapter 8). Ideally, the residues trials will generate all possible relevant analytes covering both forms of the residue definition. The selected examples in Appendix B illustrate typical scenarios encountered when the residues of a pesticide are defined. As an example, with haloxyfop-p-methyl, a rapid cleavage of the methyl ester to the free acid and conjugation in plants is observed. The residue definition includes the ester, salts and conjugates, as the available analytical method relies on hydrolysis of total haloxyfop residues and its conversion to either methyl or butyl ester and determination by GC-MS. This technique does not distinguish between R and S haloxyfop and its esters and conjugates, and therefore "any ratio" is included in the residue definition Residue definitions The residue definitions for dietary risk assessments thus should consider all residue components of toxicological interest and usually include the parent compound together with all or the main toxicologically relevant metabolites and/or degradation products (ECD, 2009a; FA, 2009a, 2012). Various factors encompassing exposure potential and relevant toxicity are considered before inclusion of a metabolite in the risk assessment residue definition. Residue definitions for dietary risk assessment at EU level are now being published per compound in EFSA Conclusion reports, and with the EU Commission published review reports associated with Annex I listing when the residue levels (MRLs, HR and STMR) have been included. For both monitoring and risk assessment, a collated source of EU residue definitions (both monitoring/enforcement and risk assessment) can be found on the website of the German Federal Institute of Risk assessment (BfR, 2009). Historically residue definitions for risk assessment have been publicly available only as either national evaluations or evaluations by the Food and Agriculture rganisation of the United Nations/World Health rganisation Joint Meeting on Pesticide Residues (FA/WH JMPR). Residue definitions for monitoring/enforcement are intended to be as simple as possible and often refer only to the active substance itself (ECD, 2009a: FA, 2009a, 2012). These can be used to indicate exceedences of the MRL of the pesticide and can be analysed and quantified easily by a broad base of national laboratories, ideally using a multi-residue method. For monitoring/enforcement, current MRLs and corresponding residue definitions at EU level are, for example, available from the EU Pesticide database 12 and Regulation (EC) No 396/2005 and its amendments Comparison of residue definitions by EFSA and JMPR Since food safety nowadays is a global issue, European Guidelines are explained in an international context in Appendix A. Within the framework of the Codex Alimentarius, the FA/WH Joint Meeting on Pesticide Residues (JMPR) is the risk assessment body responsible for establishing residue definitions. In the outsourced project on metabolic processes (AGES, 2010) it was noted that the residue definitions across active substances are not always the same, when comparing EFSA conclusions and JMPR reports. The PPR Panel screened 43 decisions on residue definitions on pesticide active substances for monitoring/enforcement and risk assessment taken by EFSA in the period of and JMPR to build on the work by AGES and to investigate reasons why the conclusions differ. It was noted that for only 14 compounds did EFSA and JMPR derive the same residue definitions for risk assessment and for 24 compounds EFSA included more metabolites in the definition than JMPR. Selected examples are presented in Appendix C. 12 EU pesticide database EFSA Journal 2012;10(07):

14 The risk assessment and residue definitions of active substances were often concluded by JMPR prior to the EFSA assessments, which may have influenced the data package submitted as well as the interpretation of the guidelines followed. Additionally, JMPR tends to consider a wider range of uses than those indicated in the Annex to Regulation (EU) No 540/ ver time the emphasis on the relevance of metabolites has increased. This is common to both EFSA and JMPR, however, a more cautious approach seems to be taken in the EU than within Codex, with an associated greater attention to metabolites present at low levels. The recently adopted ECD Guidance Document on Definition of Residue (ECD, 2009a) provides an overall framework rather than the detailed guidance required to achieve harmonised residue definitions. nce opinion follow on guidance based on the current opinion has been developed by EFSA, the PPR panel suggests that, to promote international harmonisation and since the methodologies discussed are expected to have wide applicability, the ECD are asked to consider whether inclusion of this approach could be a useful addition to their Guidance on Definition of Residues Current status and future development of analytical methods for establishing residue definitions Toxicological relevance and predicted exposure levels of residues are the primary drivers for whether a metabolite should be included in a residue definition. However, it is also often the case that analytical methods are a constraining factor in decision making on what the residue definitions should be. This is due to the physico-chemical properties of a pesticide and its various metabolites that can affect extraction/clean-up/chromatographic properties and/or detection or a combination of all of them. The choice of analytical method can be influenced by specific methodological factors, and there are a variety of reasons why an analyte or series of analytes can present particular challenges. In some circumstances, it may not be feasible to analyse a parent pesticide and the metabolites of that pesticide simultaneously, especially if the nature of a metabolite is quite different from that of the parent, for example, for triazole pesticides, the parent molecule commonly needs to be analysed by a different method than to the metabolites of interest that include triazole alanine and triazole acetic acid. It may not be possible to determine a pesticide and its metabolites separately; this can be an issue with methods that involve a derivatisation step e.g. for cycloxydim where the parent and metabolites are converted to a glutaric acid derivative, thereby using a common moiety approach. A different problem for analysis of some pesticides is that a unique hydrolysis step may be required to enable the residues to be extracted and this may have an impact on the recovery of free and conjugated residues; such a hydrolysis step would not be suitable for determination of other pesticide analytes. For those pesticides that are particularly difficult to determine, specifically adapted single residue methods or "common moiety methods" should be applied. Some common moiety approaches may further be problematic because they are not specific to the pesticide that has been applied e.g. for bisdithiocarbamates where residues are converted during analysis to CS 2. All ethyl- and propyl dithiocarbamates, including their metal complexes like mancozeb, are detected through generation of carbon disulfide and assayed by colorimetric or chromatographic methods. It is not possible to distinguish the amount of CS 2 found into parent compounds and metabolites, as this information is lost in the process. Furthermore, other precursors of CS 2 in the sample may contribute to the result. In this case it is not possible to perform a refined risk assessment since different dithiocarbamate pesticides and other substances have different toxicities. Therefore, if there is a choice, specific methods that quantify concentrations of individual compounds of related residues are preferred over common moiety methods. These common moiety methods may also not be ideal for assessing cumulative exposures since whilst they can determine some combinations of residues, the analysis masks the information on residue levels of individual active 13 Regulation (EU) No 540/2011 of 25 May 2011 implementing Regulation (EC) No 1107/2009 of the European Parliament and of the Council as regards the list of approved active substances. fficial Journal L 153, June EFSA Journal 2012;10(07):

15 substances or metabolites that would otherwise be available if specific methods for each analyte had been used. The generic requirements of an analytical method in terms of performance and to a lesser extent cost can result in compromises on the nature and level of detectable metabolites that can be included at any one time. The various difficulties described and the challenge of being able to determine a large number of pesticides simultaneously are why the international community has derived the two-fold approach in setting different residue definitions (as described in chapter 2.1) in order to circumvent this compromise, and, on occasion, allow the use of different analytical approaches to cover conflicting analytical requirements. The methods for pre-registration of active substances aim to identify and quantify the active substance and metabolites (i) to generate residues data on which consumer dietary exposure assessments are based, and (ii) to support studies on the fate and behaviour of the active substance in foodstuffs, the environment, ecotoxicology and toxicology (EC, 2000). The methods of analysis can use more complex analytical approaches if warranted in order to fully cover the scope of all the components that are in the residue definition for risk assessment. Therefore, during the development of these analytical methods the applicant aims to tailor the methods for various matrix types to all the analytes of potential interest from a risk assessment perspective. Such a tailored approach to method development is not suitable for the analytical methods for monitoring and enforcement, since the aim of these post-registration methods that are used by monitoring laboratories is to fit preferably into existing multi-residue methods, which reliably detect several hundreds of compounds in a cost effective manner. It is thus not feasible to achieve the concurrent analysis of the large number of potentially applied pesticides together with their toxicologically relevant metabolites and/or breakdown or reaction products. Therefore, the simpler monitoring and enforcement residue definition is used based on a marker concept and MRLs that are set cover only the level of residue for the analyte(s) included in the monitoring and enforcement residue definition. Applicants are requested to demonstrate, whether the components in residue definition for monitoring and enforcement can be analysed reliably using multi-residue methods. To do this, applicants may test existing published multi-residue approaches or they may justify that the method they have proposed uses commonly available laboratory equipment. Ideally, all pesticides would be determinable by multi-residue testing; however this is not always achievable and single methods for particular pesticides need to be developed and used, an approach which tends to add significant additional cost to the monitoring programmes. The herbicide glyphosate is an example of a compound which has required a single method approach; glyphosate, through its amphoteric nature (glyphosate carries both negative and positive charges at physiological ph conditions), escapes conventional extraction and clean-up schemes and has to be analysed by a sequence of tailor-made steps such as ion-exchange extraction and pre- or post-column derivatisation to achieve the necessary limits of detection. The approach to derive residue definitions for both for monitoring and risk assessment as described earlier is only really possible if a conversion factor (as a multiplication factor) can be proposed to enable the residue level determined in the monitoring to then be converted to a corresponding level for risk assessment purposes. In this way, a risk assessment can be performed covering the components of interest to risk assessment that were not actually analysed in the residue monitoring. Conversion factors are further discussed in chapter 8.5. Practical experience of working with conversion factors seems to support the view that they are not easy to set, are not available for every crop circumstance, and they are not always used (see chapter 8.5.). The latest advances in the analytical methodology (generic and simple extraction procedures with advanced mass spectrometers with liquid and/or gas chromatography) facilitate simultaneous determination of an increasing number of pesticides and metabolites, without compromising the reliability of the methods. However, conversion factors are still considered necessary because there tends to be a limitation on the total number of compounds analysable in a multi-residue method, the availability of analytical standards for metabolites, the time required to process the information and EFSA Journal 2012;10(07):

16 the costs of the analysis. Whilst technical developments and research can be harnessed to solve analytical problems and improve the reliability and efficiency of methods of analysis, issues of scale and cost will always need to be addressed. The utility of monitoring programmes in assessing dietary exposures to pesticides, not only considering compliance with MRLs, depends on the ability to evaluate as large a number of pesticide residues as possible at low levels, and to ensure that a full risk assessment covering all relevant metabolites can be performed based on the results obtained. In the future, these risk assessments will also need to cover cumulative exposures Relevant metabolites This opinion relates to Regulation (EC) No 1107/2009 which stipulates that special attention should be paid to whether the metabolite poses a higher or comparable risk to organisms than the parent substance or if it has certain toxicological properties that are considered unacceptable. The ECD Guidance document (ECD, 2009a) proposes a list of aspects to be considered when deciding on inclusion or non-inclusion in the dietary risk assessment when the different aspects presented in (Table 1) are considered. Table 1: Aspects to consider when deciding on inclusion or non-inclusion of metabolites in the dietary risk assessment (excerpt from ECD Guidance document, 2009a) More likely to be included Parent compound is highly toxic Metabolite/degradate likely to be found in commodities that are human food Metabolite/degradate levels in magnitude of residue studies exceed those expected from metabolism studies Metabolite/degradate is not formed through metabolism in rats Parent compound was non-detectable, but metabolites were found in high levels in metabolism studies Less likely to be included Parent compound has low toxicity relative to expected exposures Metabolite/degradate found in only one matrix at 10-20% of the total residue (unless that matrix is a major human food) Metabolite/degradate present at very low levels (in mg/kg) Metabolite/degradate structure is similar to innocuous chemicals Metabolite/degradate occurs predominantly in animal feeds rather than commodities that are human foods Hydrophilic metabolites less toxic than the parent compound Considerations for drinking water: Environmental degradate is persistent Environmental degradate has low soil binding potential Degradate is detected in water monitoring studies Considerations for drinking water: Environmental degradate is short-lived Environmental degradate has high soil binding potential Degradate is not detected in terrestrial field dissipation studies The ECD Guidance document does not explain in detail how to assess the toxicological burden of metabolites. This is therefore addressed in the following chapters. 3. Rationale for the outsourced projects The establishment of the residue definition for the purpose of consumer risk assessment involves a decision on which metabolites are of toxicological concern. EFSA Journal 2012;10(07):

17 As only limited information on the toxicological properties of metabolites is usually available, EFSA considers that assessment methods and alternative scientific tools, not involving animal testing, need to be evaluated. Such a consideration and/or the development of these approaches are needed to optimise the consistency and robustness of the evaluation of the toxicological profile of the metabolites from plant protection products. EFSA outsourced four activities to develop this task: The metabolic pathways and degradation processes may modify the toxicological properties of pesticide active substances. Due to the lack of experimental toxicological data on metabolites, the EU peer review of pesticides used a case by case approach in the evaluation of toxicological relevance of metabolites. Factors such as the presence of the metabolite in in vivo metabolism studies and structural similarity have been used as possible indicators of toxicity similar to parent compound. The aim of the outsourced project was the assessment of the scientific evidence, regarding the possible influence of metabolism on the toxicity of pesticides, see chapter 4. The Threshold of Toxicological Concern (TTC) approach is based on the principle of establishing a human exposure threshold value for chemicals, below which there is a very low probability of an appreciable risk to human health. The TTC approach allows identification of threshold values for chemicals of unknown toxicity considering only their structure and toxicity data on chemicals sharing broadly similar structural characteristics. The outsourced project was aimed at assessing the usefulness of the TTC approach for pesticide metabolites, while taking into account the legal framework of Directive 91/414/EEC (now Regulation (EC) No 1107/2009), see chapter 5. Applicability of (Q)SAR analyses - the basic assumption of (Quantitative) Structure Activity Relationship (Q)SAR analysis in risk assessment is that biological activity of a chemical depends on its intrinsic properties and can be directly predicted from its molecular structure and inferred from properties of similar compounds whose activities are known. The outsourced project was aimed at exploring the applicability of computational methods in the evaluation of toxicological relevance of pesticide metabolites with a focus on genotoxicity alerts, see chapter 6. Applicability of (Q)SAR analysis for toxicity and neurotoxicity - in order to refine a proposed assessment scheme for acute effects involving the TTC approach (which is based on long term effects), and with regard to identifiying metabolites that would need to be considered in this scheme, a project was outsourced to evaluate if (Q)SARs can be used to identify pesticides having and/or neurotoxic effects. Experience has shown that these are generally the critical endpoints following short term exposure, see chapter 7. It is noted that none of the outsourced projects sought to specifically address the possible impact of stereoisomerism since this part of ToR (Terms of Reference) was added at a later stage. The PPR Panel, building on the outsourced projects, the ECD guidance document, and its own research will in this opinion present approaches on how to evaluate the toxicological relevance of metabolites of pesticide active substances, including stereoisomers, in dietary risk assessment. 4. Impact of metabolic processes on the toxicological properties of pesticide residues To review current knowledge on the importance of metabolic processes to the toxicology of pesticides, the project Impact of metabolic and degradation processes on the toxicological properties of residues of pesticides in food commodities was outsourced to the Austrian Agency for Health and Food Safety (AGES). The contractor addressed two main issues related to the impact of metabolism in the evaluation of the toxicological relevance of metabolites of pesticide Active Substances for Dietary Risk Assessment: a) the criteria applied for evaluating pesticide metabolites in different regulatory contexts; b) evaluation of metabolic pathways with respect to their toxification/detoxification potential for selected groups of pesticides (AGES, 2010). EFSA Journal 2012;10(07):

18 4.1. Evaluation of toxicological profiles of pesticide metabolites in different regulatory contexts (results of outsourced project) AGES screened scientific literature and guidance documents in order to evaluate different approaches for handling metabolites on the basis of their toxicity. The criteria applied for evaluating the toxicological profiles of pesticide metabolites were also considered by analysing decisions on metabolites made by EFSA and JMPR between 2006 and The contractor noted differences between EFSA and JMPR on the inclusion of metabolites in the residue definition. It was observed that no clear criteria are available to support the decision of when the toxicity of metabolites is considered covered by the studies on the active substances, on the basis of their concentration in body fluids. Decisions are made on a case by case basis, based on expert judgement. As a conclusion of their review, AGES suggested some criteria to be applied in the evaluation of pesticide metabolites: A metabolite occurring in rat or livestock metabolism studies at > 10% in body fluids needs to be assumed as present in sufficient amounts to contribute to the overall toxicological profile, as considered for pharmaceuticals. Metabolites and their precursors/intermediates found in livestock (ruminants, poultry) should be considered as present in rat metabolism even if not measured, based on the assumption that warm blooded animals have comparable metabolism. Conjugates of metabolites found in plants should not be automatically assumed to be of no concern, since they can be cleaved to release free unconjugated metabolites. These conclusions are discussed by the PPR Panel in chapter Evaluation of metabolic pathways with respect to their toxification/detoxification potential for selected groups of pesticides as presented by the contractor. AGES analysed the potential impact of the structural changes to the parent compounds during metabolism on the toxicological properties of the derived molecules through a comparison of the toxic effects of the metabolites and parent compounds. All of the active pesticide compounds entered on CIRCA (424) were considered, and 11 chemical classes of pesticides were selected on the basis of common metabolic pathways, representativeness in the chemical group (at least four active substances), number of active substances in Annex 1 of Directive 91/414/EEC and number of new compounds. The chemical groups selected were: sulfonylureas, triazoles, aryloxyphenoxy-herbicides (FPs), chloroacetamides, strobilurines, dinitroanilines, benzimidazoles, neonicotinoids, carboxylic acids and amides, cicarboximides and macrocyclic lactons, containing 56 parent compounds and their metabolites. All the data available on the ADME studies and on the toxicological properties of the active substances and metabolites were collected and evaluated. The descriptions of the ADME studies in the evaluated DARs were very heterogeneous and not sufficiently detailed. The large majority of studies available on metabolites were acute toxicity studies giving information on LD 50 and on some clinical observations. Data on subchronic or studies were available only in a few cases. In addition, the studies on parent compounds and metabolites were carried out in different species (rat or mice) or strains, and/or in different experimental conditions. The presence of more than one metabolic step between an active compound and its metabolites in some cases impaired the attribution of a toxification potential to a specific metabolic reaction. EFSA Journal 2012;10(07):

19 Despite these constraints, and that this exercise to compare the toxicity of the metabolites and parent compounds has a high level of associated uncertainty, some general conclusions were drawn by the contractor: It is not possible to attribute a toxification/detoxification potential to a specific metabolic step, although a number of metabolic steps were identified as probably not causing higher toxicity of metabolites, as shown for several compounds (simple demethylation of the ring or side chain, simple hydroxylation of the ring system without any cleavage of the ring, hydroxylation of another ring position than the parent, conjugation of metabolite with amino acids). The metabolic pathways are in most cases specific to the chemical groups. The toxification/detoxification depends on the toxicological profile of the parent compounds: a detoxification reaction in one case could be a toxification step in another case. As a general suggestion, the contractor proposed a revision of the study design and harmonisation of criteria for the selection of radiolabel positions and of kinetic parameters for metabolites in ADME studies. The contractor also suggested the replacement of acute toxicity studies with metabolite testing using subchronic studies: the 90 days rat study ECD TG 408 (ECD, 1998) was recommended. The use of animals of the same species, strain, age and sex as used in the toxicity studies of the active substance was also recommended, in order to avoid potential differences in detoxification processes Conclusions of the PPR Panel Evaluation of metabolic pathways with respect to their toxification/detoxification potential for selected groups of pesticides. The PPR Panel considers that the outcomes of the case studies on the potential impact of structural changes to parent molecules on the toxicological profiles of derived compounds, although related to a limited number of chemical groups, reflect the inadequacy of the available database on toxicokinetics and on toxicological profiles of pesticide metabolites. The PPR Panel outlines the need for adequate toxicokinetic data which should be used to help in study design in order to improve the efficiency of toxicity testing of pesticide metabolites. The new ECD Guidance document on toxicokinetics ECD TG 417 (ECD, 2010), meets the recommended criteria: it addresses the main requirements to obtain data on absorption, distribution, metabolism, excretion and potential bioaccumulation and to provide useful information for planning and evaluating the toxicity studies and for understanding the mode of action of the compounds. Specific recommendations are made on the use of the same animal species and strain for ADME and toxicological studies on metabolites. Studies on the possible effects on enzyme induction/inhibition are considered. In line with TG 417, the PPR Panel also recommends the use of physiologically-based pharmacokinetic (PBPK) modeling as an approach to be considered in the assessment of ADME processes. PBPK models are quantitative descriptions of the absorption, distribution, metabolism and excretion (ADME) of chemicals in biota based on interrelationships among key physiological, biochemical and physicochemical determinants of these processes. PBPK models are applied in pharmaceutical research and health risk assessment in facilitating the prediction of inter-individual, interspecies and route-to-route differences in dose metrics based on physiological and physicochemical properties. ADME processes usually have considerable complexity where physiological, physico-chemical and metabolic processes contribute to the fate of a compound in the organism. PBPK models tend to EFSA Journal 2012;10(07):

20 simplify and separate the processes in a multi-compartment model, where the compartments represent predefined organs or tissues (or groups of these) connected by a stream of body fluids. Recently a guidance document on "Principles of Characterizing and Applying PBPK Models in Risk Assessment" has been developed within the framework of the IPCS project on the harmonisation of approaches to the assessment of risk from exposure to chemicals (IPCS, 2010) Considerations by the PPR Panel on conjugated and bound residues The detoxification mechanisms of exogenous organic compounds in higher plants lead to different products, conjugates and bound (also known as non-extractable) residues, with different toxicological properties. Pesticide conjugate metabolites are formed by reaction of parent compounds or their phase I metabolites with endogenous substrates like glutathione, sulphate or sugars. In general, some conjugates can readily be cleaved. Due to possible release of unconjugated product, conjugates should be considered as potentially bioavailable as their unconjugated products. As a consequence they should be included in the evaluation of toxicological relevance. The number of conjugate types is extensive, the most frequently reported being glycosides, sulphates and metabolites resulting from glutathione conjugation. Limited information is available on the stability of conjugates to enzymatic and hydrolytical attack in the gastrointestinal tract of humans and livestock species (DEFRA, 2007). The outcomes of a project carried out with twelve different β-dglucosides, as the most common type of conjugate, confirmed the differences in behaviour related to the different functional chemical groups and glycosidic linkages (DEFRA, 2007), suggesting that the evaluation of toxicological relevance should be done on a case-by-case basis. The situation with bound residues is more complex than for conjugates. Bound residues covers a range of molecules: covalently bound but otherwise intact metabolites; metabolites that are physically encapsulated within the macromolecular matrix of plant and animal tissues; transitory simple molecules, such as malate, pyruvate and others which are used to produce the plethora of biomolecules in the plant. An approach to distinguish the covalently bound metabolites and the endogenous labeled biomolecules that have become naturally incorporated has to be found. Whilst these types are considered bound residues, it is generally agreed (Sanderman, 2004) that only the covalently bound or physically encapsulated but intact molecules are of toxicological concern. The criterion to apportion the total radioactive residue (TRR) which is unextractable between the randomly labeled biomolecules and the other bound metabolites is therefore to use hydrolytic or enzymatic cleavage to liberate the metabolites. Where no metabolite can be cleaved, the residual TRR tends to be considered as not bioavailable and is not taken into account for toxicological relevance. The data generated by more sophisticated studies (DEFRA, 2007) suggests that routine chemical approaches to release bound residues from unextractable TRR may over-estimate bioavailability of bound residues. It is recognised that the available research on bioavailability of xenobiotics is currently limited Evaluation of toxicological profiles of pesticide metabolites. The PPR Panel, taking into account the comments/suggestions made by the contractor on the criteria for evaluating toxicological profiles of metabolites as well as the current approach applied by EFSA in peer review of pesticides, concludes that: EFSA Journal 2012;10(07):

21 EFSA decisions on the toxicity of pesticide metabolites should be made on a case by case basis considering different factors and not only the concentration of metabolites in body fluids. Due to the wide heterogeneity of available data it is difficult to establish a standardised procedure in the evaluation of relevance of metabolites to the overall toxicological profile. A pragmatic approach needs to be followed, on a case-by-case basis, taking into account several factors such as the presence of the metabolite as an intermediate in the rodent metabolic pathway, the similarity of the chemical structure of the metabolite and the parent compound, and the structural similarity of the metabolite to known classes of toxic compounds. The data on livestock metabolites have mainly qualitative significance considering the restricted number of tested animals: a ruminant metabolism study can be carried out on a single animal. The metabolism studies in livestock are carried out under different exposure conditions and using different dose ranges than rat studies. Additionally, differences in metabolic pathways and in the extent of metabolism can be envisaged in different species. The information on the intermediates in the metabolic pathway of the rodent could be of importance in the evaluation of the relevance of the livestock metabolites, and this should be done case-by-case using a weight of evidence approach. The PPR Panel considers that an improved understanding of the behaviour of pesticide conjugates in the gastrointestinal tract of human and livestock species is needed, e.g. the gut of ruminants with its specialised stomach anatomy and physiology, and associated microflora is quite different to that of rats and humans, and the impact of this on the dietary risk assessment could benefit from further study. The PPR Panel considers that a case-by-case approach to the evaluation of conjugates and bound residues needs to be adopted, that takes account of the experimental methods used in investigating these types of residues. Reasonable effort should be made in terms of methodological approaches to demonstrate that residues are bound prior to concluding that they are not of relevance to the risk assessment. Where metabolites are found as conjugates, it is usually reasonable to conclude that they are bioavailable and the estimation of exposure to metabolite should be based on the sum of the free and conjugated residues. If specific evidence is available to support a different approach for conjugates, then this should be assessed on its scientific merit. 5. Threshold of Toxicological Concern (TTC) The Threshold of Toxicological Concern (TTC) approach was considered by the PPR Panel, as a tool for providing scientific advice about possible human health risks from low levels of exposure to metabolites of pesticides. The TTC approach is also the subject of a separate opinion from the EFSA Scientific Committee (EFSA, 2012) related to the relevance and reliability of the TTC concept for application more general application in the food and feed area. The TTC approach is based on a fundamental principle of toxicology, that toxicity depends upon dose and duration of exposure 14.The TTC is based on a number of generic structural and other characteristics, where exposure threshold values could be established for chemicals below which there is no appreciable risk to human health. The TTC approach therefore enables the identification of the relevant threshold value for a chemical of unknown toxicity considering only its structure and a toxicity database on chemicals sharing broadly similar structural characteristics. It should be noted that the TTC exposure values are derived using a probabilistic approach. 14 For an explanation on the ongoing discussions regarding low-dose toxicity, see chapter 5.4. EFSA Journal 2012;10(07):

22 The first scheme for predicting the hazard of chemicals based on their structure was developed by Cramer et al., (1978). Three different chemical classes were proposed, with increasing order of oral toxicity, based on whether the compound is a normal constituent of the human body, its potential reactivity and the nature of functional groups present in the molecule. A decision tree using a series of structure-related questions was applied to allocate the chemicals into one of the three classes. The logic of the sequential questions was based on contemporary knowledge on toxicity and on metabolic pathways in different mammalian species. The Cramer classification scheme was further developed and revised following extensive analyses of available chronic oral toxicity data (Munro 1990; Munro et al., 1999; Kroes et al., 2004). The distribution of NELs (No bserved Effect Levels) for chronic effects were plotted for each Cramer class and the 5 th percentile of each distribution was used to derive a threshold value, with a 95% probability that the NEL of an unstudied compound allocated to the same class would be higher than the value derived. TTC values were obtained for the Cramer classes by dividing the 5 th percentile NELs by the default uncertainty factor of 100 (to give TTC values of 30, 9, 1.5 µg/kg bw/day) and multiplying by a default body weight of 60 kg (1800, 540 and 90 µg/person/day for classes I, II and III respectively). Regarding neurotoxicity, the overall distribution of NELs for organophosphates (Ps), the most potent compounds in the neurotoxicity database, was around one order of magnitude from the distribution of NELs for other compounds in Cramer class III. A TTC value for P compounds (0.3 µg/kg bw/day or 18 µg/person/day) was proposed and a step to identify a structural alert for Ps was introduced into the decision tree (Kroes et al., 2004). The TTC values for the Cramer classes were developed for application to chemicals with no structural alerts for genotoxicity. A TTC value for substances with a structural alert for genotoxicity (0.15 µg/person/day or µg/kg bw/day) was established by linear extrapolation from the TD50 values obtained from animal cancer studies to a risk of 1 in 10 6, through the analysis of an expanded version of the Carcinogenic Potency Database (CPDB, US FDA, 1995; Gold and Zeiger, 1997) using the results from genotoxicity tests and structural alerts for genotoxicity. A number of high potency genotoxic compounds (aflatoxin-like, N- nitroso-, azoxy-compounds), as well as certain other very potent compounds (steroids, polyhalogenated dibenzo-p-dioxins and dibenzofurans) may still be of concern at the TTC of 0.15 µg/person/day and were therefore excluded from the TTC scheme (Cheeseman et al., 1999; Kroes et al., 2004). All metals and inorganic chemicals were also excluded since they were not covered by the database. The science behind the TTC approach was critically examined by the EFSA Scientific Committee (EFSA, 2012) in order to evaluate whether the human exposure threshold values, for cancer and noncancer endpoints, are sufficiently conservative. The information on the toxicological database sources used, and the types of endpoints that determined the NELs were considered. In addition an assessment of the original published papers and reports referenced in the database on the substances in the lowest 10 th percentile of the distribution of NELs for classes I and III was carried out, in order to assess the quality of the studies and whether the NELs identified were appropriate. The EFSA Scientific Committee concludes that, where a conservative estimate on human exposure is available, the TTC values for compounds with genotoxic alerts, with anti-cholinesterase activity, and compounds classified in Cramer class I and III are sufficiently conservative to be applied in risk assessment of substances of unknown toxicity present at low levels in food. The TTC value for Cramer class II substances, derived from toxicological data on only a few compounds, is not well supported by the presently available databases. The suggestion of the EFSA Scientific Committee is to treat substances that would be classified in Cramer Class II as if they were Cramer Class III substances. The EFSA Scientific Committee, following the analysis of substances classified for reproductive and toxicity under the EU legislation, also considered the TTC values for Cramer class I and III sufficiently protective for these effects (EFSA, 2012). EFSA Journal 2012;10(07):

23 The EFSA Scientific Committee considered the situation with regard to substances that may have endocrine-mediated toxicity since the TTC approach might not be applicable to such substances due to uncertainty about low-dose effects (Kroes et al., 2004; Cheeseman et al.,1999). The Scientific Committee also identified steroids as a group that includes some potent carcinogens. The Scientific Committee overall concludes (details in footnote 15 ) that in most situations where the TTC approach could be applied there would be no a priori knowledge that a substance has endocrine mediated activity. The Scientific Committee recommends that if there are data showing that a substance has endocrine mediated toxicity, then the risk assessment should be based on those data, rather than using the TTC approach, as would be the case for adverse data on any other endpoint. The SC TTC opinion does not provide any specific recommendations on which computational genotoxicity tool could be used. The Scientific Committee is not confident about the general applicability of available proposals for adjusting TTC values for short term exposure to substances with structural alert for genotoxicity. It is recommended to address the issue of less than chronic exposure case-by-case by considering the margin between the appropriate TTC value (without any adjustment for duration of exposure) and the estimated dietary exposure. The Scientific Committee concluded that the TTC approach should not be used for the following (categories of) substances: high potency carcinogens (i.e. aflatoxin-like, azoxy- or N-nitrosocompounds, benzidines, hydrazines), inorganic substances, metals and organometallics, proteins, steroids, substances with a high potential for bioaccumulation, nanomaterials, radioactive compounds, and mixtures of substances containing unknown chemical structures. A revision and refinement of the TTC scheme, based on the updated databases and on the advances in knowledge is recommended for the future. The EFSA Scientific Committee opinion on TTC concludes that the TTC values provide adequate assurance of protection of sensitive subpopulations, with the exception of young infants under the age of 6 months, where careful consideration in the application of TTC is needed. The TTC values should be expressed in µg/kg bw/day for comparison with exposure for the respective age groups. The Scientific Committee noted that the current TTC values for non-cancer endpoints are applicable to chronic exposure, as, with the exception of the TTC value for organophosphate and carbamate structure, they were derived from databases that do not address effects from acute exposure. The 15 From EFSA 2012: Intensive discussions are also taking place within the European Union under the aegis of the Community Strategy for Endocrine Disrupters, which is addressing the key requirements of further research, international cooperation, communication to the public, and appropriate policy action. A draft of the measures concerning specific scientific criteria for the determination of endocrine disrupting properties in relation to human health impacts is anticipated to be ready by the end of These measures are required, in particular, for the legislation governing REACH and the Plant Protection Products Regulation, but the intention is to develop a systematic approach for the identification and assessment of endocrine disruptors which can be applied across the different pieces of EU legislation. The general concept should be consistent and should ensure that endocrine disruptors are dealt with in a consistent and co-ordinated manner across the EU (EC, 2011a). Regarding the issue of substances that may have endocrine-mediated toxicity, the Scientific Committee concludes as follows: a. In most situations where the TTC approach might be applied, there would be no a priori knowledge that a substance has endocrine activity. b. If there are data showing that a substance has endocrine activity, but the human relevance is unclear, then these data should be taken into consideration, case-by-case, in deciding whether or not to apply the TTC approach. c. If there are data showing that a substance has endocrine-mediated adverse effects, then, as would be the case for adverse data on any other endpoint, the risk assessment should be based on the data, rather than the TTC approach. d. In view of the extensive work, currently ongoing, to develop an EU-wide approach for defining and assessing endocrine disrupters, once that approach is finalised it will be necessary to consider any impact it may have on the use of TTC approach. e. In the meantime, the Scientific Committee recommends that untested substances, other than steroids, can be evaluated using the TTC approach recommended in this opinion. EFSA Journal 2012;10(07):

24 Scientific Committee considered that there is insufficient information to enable any recommendations for a reliable/appropriate means of adjusting the TTC values for shorter durations of exposure. However, there is no scientific reason why this should not be possible, with a suitable database. In cases where TTC values are exceeded the use of a refined approach for exposure assessment and/or chemical specific toxicity data on a case-by-case basis was recommended. A software package 16 is available to estimate toxic hazard of chemicals by applying a decision tree approach, including the original Cramer rulebase, with extensions, and the TTC decision tree of Kroes et al., (2004). The Scientific Committee recommends that when the TTC approach is applied to substances with closely related structures and to which there is co-exposure, it may be appropriate to sum their exposures, as would be done in a cumulative risk assessment on substances with the same mode of action Current applications of TTC concept The PPR Panel reviewed the current fields of application of the TTC approach in different regulatory contexts Additives: JECFA and EFSA The TTC approach was adopted by the Joint FA/WH Expert Committee on Food Additives (JECFA) to evaluate flavouring substances in 1997 (FA/WH, 1997) and has since been modified several times (FA/WH, 1999, 2006, 2009). The JECFA decision tree scheme places chemicals into Cramer structural classes and then makes decisions on the need for toxicity data based on whether or not intakes under the expected conditions of use will exceed the threshold of toxicological concern for the relevant structural class. A further threshold (1.5 μg/person/day US FDA threshold of regulation based on carcinogenic risk) is applied at the final step of the scheme for substances for which no toxicity data are available to provide a threshold with an adequate margin of safety. If this 1.5μg/person/day threshold is not exceeded by the estimated intake, then it is concluded that no data are required for such substances (which have passed earlier steps in the decision tree), provided that they do not contain structural alerts for genotoxicity. An evaluation of the data on the application of the TTC approach between 1999 and 2006 on approximately 1800 flavouring substances allowed JECFA (FA/WH, 2006) to confirm the applicability of the TTC approach for flavouring agents and also to consider its application to other substances present in the diet in small amounts. It was emphasised that the TTC approach should be used only in conjunction with a conservative estimate of dietary exposure. The same procedure for the evaluation of flavouring substances was adopted by the EU Scientific Committee on Food (SCF, 1999) and has subsequently been used by EFSA in modified form since 2004 for the evaluation of several thousand substances on the EU Register of Flavouring Substances Pharmaceuticals: European Medicines Agency (EMA) The European Medicines Agency (EMA) Committee for Medicinal Products for Human Use (CHMP) proposes the use of a threshold of toxicological concern (TTC) for genotoxic impurities (EMA, 2006). The TTC refers to a threshold exposure level to compounds that does not pose a significant risk for carcinogenicity or other toxic effects. The EMA Guideline recommends a TTC of 1.5 μg per person per day for all but a highly potent subset of compounds (aflatoxin-like, N-nitroso-, azoxy-, stereoids, polyhalogenated dibenzo-p-dioxins and dibenzofurans). This threshold corresponds to an incremental 1 in 10 5 lifetime risk of cancer, a risk level that the EMA considers justified because of the benefits derived from pharmaceuticals. 16 TXTREE version August 2011 Ideaconsult Ltd., Bulgaria available via the EC Joint Research Centre website at EFSA Journal 2012;10(07):

25 The Guideline indicates that a TTC value higher than 1.5 μg per day may be acceptable in situations where the anticipated human exposure will be short-term, for the treatment of life-threatening conditions, when life expectancy is less than 5 years, or where the impurity is a known substance and human exposure will be much greater from other sources. This is based on a weight-of-evidence approach taking account of the profile of genotoxicity results. A reduction by a factor 10 was proposed for the acceptable daily intake of genotoxic impurities for short term exposure in pediatric and young adult patients. The acceptable limits for daily intake of genotoxic impurities are 5, 10, 20, and 60μg/day for duration of exposure of 6-12 months, 3-6 months, 1-3 months, and less than 1 month, respectively. For a single dose, an intake of up to 120 μg is acceptable. When more than one genotoxic impurity is present in the drug substance, the TTC value of 1.5 μg/day can be applied to each individual impurity provided the impurities are structurally unrelated. In case of structural similarity, it can be assumed that the impurities act by the same genotoxic mode of action and have the same molecular target and thus might exert effects in an additive manner. In such a situation, the sum of the genotoxic impurities limited to 1.5 μg/day is recommended (Muller et al., 2006). Genotoxicity testing is not obligatory when a potential genotoxic impurity is controlled at the TTC level, or if the testing batch of drug substance with the impurity at a level of 0.05% is negative in a genotoxicity battery. The Guidelines recommend an expert scientific review of the synthetic route and the chemical reactions and conditions involved to identify compounds of special concern. This review should include an evaluation of structure-activity relationships (SAR) for genotoxicity. The absence of structural alerts based on a well-performed assessment (e.g. through the application of commonly used SAR assessment software including DEREK and MCASE) is sufficient to conclude that the impurity is of no concern with respect to genotoxicity. Compounds showing positive alerts not present in the active substance need to be tested with a bacterial gene mutation test. A negative bacterial gene mutation test overrules the structural alert (EMA, 2010) Pesticide metabolites in groundwater Guidance document SANC/221/2000 rev 10 of 25 February 2003 (EC, 2003) The strategy for the assessment of the relevance of pesticide metabolites in ground water as described in the guidance document includes a TTC approach for those metabolites that are considered not relevant (in the sense of point C of the annex VI of the Directive 91/414/EEC and of the Directive 98/83/EC 18 regulating the quality of water intended for human consumption) after tiered hazard screening steps. As a pragmatic approach to limit the requirement for animal tests, it is accepted that such non-relevant metabolites can be present in groundwater up to a concentration of 0.75 µg/l. Assuming a consumption of 2 liters of water per day, this corresponds to the threshold of 1.5 µg/person/day set by the US FDA for the Threshold of Regulation Industrial chemicals (REACH): ECHA Some application of the TTC approach is envisaged in the REACH regulation, but limited to cases where there are only a few exposure scenarios that are well characterised. The REACH regulation states the need for non-testing methods and considers the possibility of waiving tests on the basis of exposure considerations. A TTC value could be applied to waive some specific tests (e.g. repeated dose or reproductive toxicity) where it can be demonstrated that there is no significant human exposure Applicability of TTC approach in the dietary risk assessment of pesticide metabolites: outcome of the outsourced project. Within the frame of a commissioned project Applicability of thresholds of toxicological concern in the dietary risk assessment of metabolites, degradation and reaction products of pesticides (CRD, 18 Council Directive 98/83/EC of 3 November 1998 on the quality of water intended for human consumption. fficial Journal L 330, December EFSA Journal 2012;10(07):

26 2010), the contractor Chemicals Regulation Directorate (CRD) UK considered the strengths and weaknesses of the TTC scheme (Kroes et al., 2004) for it application to pesticide metabolites. In order to validate the existing TTC values for use with pesticide metabolites the TTC values should ideally be compared against experimental data for pesticide metabolites. However, for most PPP metabolites the available toxicity database is limited and it is unlikely that sufficient toxicity data would ever be available for metabolites and transformation products to adequately validate the TTC values in this way. The choice was to use pesticide active substances for which a comprehensive database is available to derive robust ADI values. It was felt that this approach was justified given the similarity of chemical structures and of potential toxic effects. 100 active substances were selected at random from a list of 500 compounds that were evaluated under the Directive 91/414/EEC. The list included a mixture of chemical classes covering existing and new active substances. ADI values were identified for each substance. The range of ADIs covered four orders of magnitude, from >1 mg/kg bw/d to mg/kg bw/d. Cramer classifications were determined for each of the 100 active substances by importing the chemical structure into a software package loaded with the Cramer rules decision tree 19. A (Q)SAR system (DEREK Version 11) was used to generate predictions for genotoxicity for each of the 100 active substances. The results showed that the overall reliability of the DEREK prediction was low. A key concern was that genotoxicity alerts were not triggered for 12 compounds with evidence of positive results in different in vitro genotoxicity assays. The effect of using a different program to predict genotoxicity was investigated applying the modules in Toxtree. The level of predictivity for the Toxtree software is not better than DEREK and a relatively poor concordance between the two programs was also observed. Combining DEREK with Toxtree did not improve the predictivity. Considering that the EU criteria for classifying a compound as genotoxic require positive results in vivo, only 5 chemicals of the 100 active substances selected for the validation exercise based on in vitro results had positive results in in vivo tests and for all an alert of some genotoxic event was triggered in DEREK (even if not necessarily on the same endpoint), although not matching the specific endpoint described in the individual studies. n the basis of these results, the TTC validation exercise was carried out following a tiered approach considering two steps before the application of the Toxtree software. 1st Step Genotoxicity: Any genotoxicity alert in DEREK resulted in allocation of a threshold of μg/kg bw (0.15 μg/person/day). 2th Step Neurotoxicity: For this exercise a neurotoxicity trigger was set for all cholinesterase inhibitors (Ps and N-methylcarbamates), compounds acting on the sodium channel (pyrethroids) and other insecticides with a structure associated with a neurological mode of action, such as neonicotinoids. For substances meeting the criteria for this trigger the TTC threshold would be 0.3 μg /kg bw (18 μg/person/day). Substances that did not reach the genotoxicity and neurotoxicity triggers were allocated to a Cramer Class using the Toxtree software. The TTC exercise with 100 active substances resulted in allocation of 95 compounds to Cramer class III, 2 to Cramer class II and 1 to Cramer class I. The TTC approach was protective compared with the ADI for 96/100 compounds. For 25/100 substances, the ADI that had been established was more than 19 (Toxtree version 1.51, Ideaconsult Ltd., Bulgaria available via the EC Joint Research Centre website at EFSA Journal 2012;10(07):

27 1000-times greater than the TTC threshold which would have applied to that substance. nly 4 active substances in the dataset of 100 had ADIs below the respective TTC value identified (Table 2). Table 2: Comparison of ADI and TTCs Genotoxicity alert Neurotoxicity alert Cramer class III Cramer class II Cramer class I TTC Threshold (μg/person/d) TTC Threshold (μg/kg bw/d) No. of substances with an ADI below applicable TTC threshold Total no. of substances considered = Compounds (Ratio: ADI/TTC) Aviglycin (0.67) Haloxyfop-R (0.43) Amitrole (0.67) MCP is a gas and deriving the ADI involved many assumptions and uncertainties (0.1) The ADIs for two compounds were established incorporating additional uncertainty factors (due to limitations in their databases) and the other two substances had a ten-fold gap between the NAEL and the LAEL in the study used to establish their ADI. In addition 1-MCP is a gas and the application of the TTC approach was probably inappropriate. The ADI/TTC ratio for the other three compounds ranged from , which, in view of the discussion above, is considered inconsequential given the overall level of uncertainty involved Validation of the developed TTC concept: case studies on metabolites A modified TTC approach for application to metabolites of plant protection products was proposed by CRD. The adaptation of the TTC concept involved two aspects: Genotoxicity: Any compound with a structural alert for genotoxicity was considered to be a potential genotoxic carcinogen and was allocated a TTC of 0.15 μg/person/day ( μg/kg bw/d). Although this clearly gives false positives, its use as a screening approach was considered appropriate. For any compound with predicted exposures above the threshold further considerations are needed Neurotoxicity: Considering that ADIs for some neurotoxic compounds are in the range of the low ADIs usually set for Ps, it was initially proposed to include metabolites and transformation products of all active substances with a neurotoxic mode of pesticidal action in the P TTC of 0.3μg/kg bw/day. Three groups of compounds were evaluated in more detail: N-methylcarbamates, pyrethroids and neonicotinoids. Like Ps the N-methylcarbamates are cholinesterase inhibitors but have a more transient effect. ADIs for two N-methylcarbamates are below the non-neurotoxic TTC value for Cramer class III of 1.5 μg/kg bw/day both oxamyl and carbofuran have ADIs of 1 μg/kg bw/day. It was therefore decided to include metabolites and transformation products of N-methylcarbamates in the neurotoxic TTC grouping. EFSA Journal 2012;10(07):

28 Data on pyrethroids and neonicotinoids were analysed separately on the basis of the derivation of the reference doses from neurotoxic (mainly tremors in dogs) or non-neurotoxic end-points and taking into account the presence of an α-cyano group. None of the pyrethroids or neonicotinoids have ADIs that would not be covered by the Cramer class III TTC value of 1.5 μg/kg bw/day. verall, there is no reason to include pyrethroids or neonicotinoids in the neurotoxic TTC grouping. The TTC approach developed was evaluated carrying out 15 case studies. The compounds for the case studies were selected based on the availability of toxicity data for their metabolites and with the intention of covering a representative range of pesticides that have been evaluated under Directive 91/414/EEC and of the possible scenarios: Few metabolites - predominant residue is parent; Few metabolites - predominant residue is not parent; Many metabolites; Profile of metabolites changes with Pre-Harvest Interval (PHI); Profile of metabolites changes with crop; Novel metabolites seen in animal transfer studies; Active substances of low, medium and high toxicity Estimation of the exposure Suitable estimates of exposure levels of metabolites need to be derived, to compare against the established thresholds of toxicological concern, in order to determine whether the need to consider a metabolite further from a toxicological perspective can be ruled out (i.e. on the basis of structure and exposure levels alone). Since, for a large number of pesticides, the levels of metabolites are not analysed directly in quantitative residues trials, the contractor proposed an approach for estimation of metabolite levels that made best use of the data available in the residues regulatory data package. Metabolism data for representative crops were taken from the Draft Assessment Reports (DAR) for each active substance. For each commodity/harvest interval/application rate considered (or combinations thereof) the ratio between the level of each metabolite (mg/kg or % Total Radioactive Residues [TRR]) and the parent compound in each of the plant metabolism studies was determined. The supervised trials median residue (STMR) levels for the parent compound were determined from the available residues trials data, conducted according to Good Agricultural Practice (GAP) supporting a particular crop use. Reference was made to DAR and FA/WH Joint Meeting on Pesticide Residues (JMPR) evaluations. The STMR for each metabolite was then determined using the median level of parent compound found in the trials and the expected ratio of metabolite to parent from the relevant metabolism studies. Long-term (chronic) intakes (NEDIs) for ten UK consumer groups were calculated using high level (97.5 th percentile) rather than average consumption data based on long term consumption patterns and the median residue found in a food commodity. Since the UK model allows intakes to be calculated for ten different consumer groups within the UK population, intakes were reported for the adult consumer group and the critical consumer group only (i.e. the group of consumers that gave the highest intake for a particular crop/residue combination) Results and conclusions of TTC case study presented by the contractor None of the metabolites had structures identified as belonging to classes of compounds of special concern that cannot be considered using a TTC approach (e.g. dioxins, metal containing or N-nitroso compounds). The metabolites were allocated TTC categories based on the following criteria: EFSA Journal 2012;10(07):

29 Assume Cramer Class III (with a TTC of 1.5 μg/kg bw/day) unless there were data/information to the contrary; DEREK predictions for genotoxicity (negative or positive) were assumed to be reliable and any alert would result in a TTC of μg/kg bw/day unless there were sound reasons to ignore the alert (e.g. test data or the identical alert was triggered in a related compound which had negative data); Metabolites of the cholinesterase inhibitors in the group of 15 were allocated a TTC of 0.3 μg/kg bw/day unless there were data to show the compound was not a potent cholinesterase inhibitor; Exposure estimates were compared with the allocated TTC. If the highest estimated exposure for a consumer group was below the TTC, the transformation product was considered to be not relevant. If one or more of the estimates was above the applicable TTC, the transformation product was considered potentially relevant and would merit further consideration. ut of a total 79 metabolites, 63 were considered non-relevant as exposure estimates were below the allocated TTC. f the 15 active substances considered, 9 had one or more metabolites that exceeded the allocated TTC. f the 16 metabolites that exceeded the allocated TTC, 9 compounds had a structural alert for genotoxicity. For the 7 remaining compounds (9%), the data available did not allow further considerations. The applied TTC scheme appears appropriate for the assessment of metabolites, degradation and reaction products. Two critical issues were identified in the application of the TTC scheme: The identification of structural alerts for genotoxicity is the first step in the TTC scheme. The software tools applied in the TTC case studies (DEREK and Toxtree) showed poor predictivity for genotoxicity, giving both false positives and false negatives, when the outcome of the analysis was compared with the available experimental genotoxicity data. A neurotoxic metabolite arising from a parent compound lacking a structural alert for neurotoxicity (e.g. organophosphate) would not be covered by the proposed scheme Conclusions of the PPR Panel on the TTC case study presented by the contractor The PPR Panel concludes that the TTC approach seems to be widely held as scientifically valid in all regulatory areas where it has been considered, as a tool for providing scientific advice about possible human risk for low level exposure. The application of the TTC approach is dependent on the quality of the underlying database and on an estimate of human exposure to the chemical in the field of application, for which there is confidence that it is not an underestimate. Where TTC approaches have not been accepted for regulatory purposes it was generally because of concerns that the databases used to derive the thresholds do not adequately cover the classes of chemicals under consideration. The TTC approach has been applied to only a limited extent in the toxicological evaluation of pesticides and their metabolites, although recent chemoinformatic analyses of TTC datasets reported in an EFSA funded study (Bassan et al., 2011), showed that the TTC databases are adequately representative of the different pesticide classes and confirmed their potential use in risk assessment. The PPR Panel considers that the TTC scheme proposed in the CRD report, as a result of a validation process and of specific case studies, would be a good starting point to develop a decision tree for the evaluation of the toxicological relevance of metabolites. The critical steps identified in the TTC scheme were considered by the PPR Panel. EFSA Journal 2012;10(07):

30 Genotoxicity prediction The first step in the TTC decision tree involves the assessment of genotoxic potential. The validation exercise performed in the TTC outsourced project applied non-testing tools in the evaluation of genotoxicity alerts. Two (Q)SAR tools (DEREK and Toxtree) applied alone or in combination showed a low predictive performance. However, this exercise suffered from the small datasets used and the heterogeneity of the available genotoxicity data. The predictability of genotoxicity using software tools was further explored with the outsourced (Q)SAR project, extending the analysis to more software tools and to more extensive datasets (see chapter 6) TTC for neurotoxicity Neurotoxic compounds were subject to a specific consideration in the CRD project. Compounds, other than Ps, with structural alerts for a neurotoxic mode of action (i.e. N-methylcarbamates, pyrethroids, neonicotinoids) were evaluated in detail. For a few compounds belonging to the methylcarbamate chemical class the ADI was below the TTC for Cramer class III. The PPR Panel therefore decided to include metabolites of N-methylcarbamates in the neurotoxic TTC grouping. Although to date no examples are known of neurotoxic metabolites arising from non-neurotoxic pesticides, the possibility that this might occur cannot be excluded. It is important to note that these metabolites might not be covered by the proposed TTC approach, unless the toxicophore formed during metabolism has already been characterised. The use of (Q)SAR tools, grouping and read-across approaches in identifying neurotoxic effects of pesticides was further addressed in an ad hoc study (see chapter 7) Application of TTC approach for acute exposure The TTC approach was designed to be applied in risk assessment of chronic exposure: the TTC values were derived from chronic studies and were based on the assumption of continuous lifetime exposure. Research on residues of pesticides in individual fruits and vegetables revealed random occurrences of comparatively high residue levels. Some individuals who consume significant amounts of such foods will occasionally eat the hot commodity unit, but this will occur only infrequently (Hamilton and Crossley, 2004). This gave rise the need to consider an approach for acute exposure assessment. The TTC concept could in principle be suitable for these situations as well. However, when using the chronic TTC values to assess acute dietary risk (which may be overly conservative, see EFSA, 2012), estimates of short term intake in many cases exceeded these TTC values. This was confirmed also for pesticide metabolites in a number of ad hoc studies carried out by the PPR Panel (see chapter 8 and Appendices E) Derivation of acute exposure thresholds The estimates of metabolite exposure performed in the context of the current opinion by way of case studies (see Appendix E, also discussed in chapter 8), expanded the work the CRD contractor had done in the outsourced TTC project on chronic exposure metabolite estimation, by also considering the potential for metabolite levels as a result of acute exposure. Since the chronic TTCs could be considered to be overly conservative for assessment of acute exposure (EFSA, 2012) it was concluded that if both chronic and acute exposure estimates for metabolites were relatively low, and below the chronic TTC thresholds, it could be proposed that no further toxicological assessment of the metabolites would be needed. In this way a screen using the chronic TTC values would be adequate to propose an assessment scheme by comparing all intake values calculated for metabolites with the TTC values. However, the case studies on metabolite estimations showed that estimations were markedly higher for the acute exposure assessments (in the PPR Panel case studies, Appendix E and discussed below) which necessitated a more specific approach for acute considerations in a TTC assessment scheme. Due to the different parameters in the metabolite estimations, it was not possible to derive a consistent factor between an acute and chronic exposure result for a particular metabolite. EFSA Journal 2012;10(07):

31 For example, for dimethoate the acute metabolite estimates (for the critical consumer) were x higher than the corresponding chronic metabolite estimate intakes, whereas for azoxystrobin the acute metabolite estimates (for the critical consumer) were x higher than the corresponding chronic metabolite estimate intakes. A difference in acute and chronic exposure of at least an order of magnitude can be realistically expected. This therefore triggered the need for extension of the TTC approach to cover the acute exposure situation for pesticide metabolites, in the scope of this opinion. In order to tackle the issue of acute exposures to pesticide metabolites, acute TTC values for the assessment of pesticide metabolites were derived from pesticide NAELs used for deriving ARfDs (i.e. short term NAELs). As a first step in the process, pesticide active substances for which an ARfD had been established were extracted from the EFSA MRL database (status 6 th May ). In this internal EFSA database all pesticides for which dietary reference values (ADI and ARfD) have been established are listed. For each substance, the information provided includes the value of the ADI/ARfD, uncertainty factor applied, source of reference value (e.g. EFSA, JMPR, etc.), year of decision, functional category (e.g. insecticide, herbicide, etc.). In total, 406 ARfDs were established for 267 different active substances. Since for several active substances there was more than one ARfD, as a second step the value considered as most relevant was selected using the following priority: EFSA (PRAPeR), Commission (Standing Committee on the Food Chain and Animal Health (SCoFCAH), European Community Co-rdination (ECC)), JMPR, Draft Assessment Report (DAR), EU Member State, so that there was only one ARfD for each active substance. For identical sources the most recent entry was selected. Substances for which a LAEL had been used to establish the ARfD (in total 5 substances) were excluded from the analysis, as the TTC approach should be based on NAELs only. Substances classified as genotoxic (i.e. all categories of germ cell mutagenicity) according to Regulation (EC) No 1272/ were excluded (in total 4 substances). As a result, 258 substances remained for further analyses. Following the Kroes scheme that provides specific thresholds for substances having a structural alert suggesting neurotoxicity, 41 substances belonging to the chemical classes of either carbamates or organophosphates were analysed separately from the remaining 217 active substances. The SMILES codes of the 217 non-neurotoxic substances were inserted in ToxTree v.2.10 for allocation into the three Cramer classes. Almost all of the 215 were assigned to Cramer Class III (2 were assigned to Cramer Class II). Based on these results and on the recommendations in the scientific opinion of the Scientific Committee on application of TTC (EFSA, 2012) questioning the adequacy of the database for Cramer Class II compounds, it was concluded that only one acute TTC threshold was necessary for pesticide metabolites without a structural alert for neurotoxicity. The distributions of NAELs for the two groups of compounds were used to identify the respective 5 th percentile values (i.e. the points on the distributions where 5% of substances had lower NAELs and 95% had higher NAELs) (see tables 1 and 2 in Appendix F). The 5 th percentile NAELs where then divided by 100 (i.e. applying the default safety factor) to give acute exposure thresholds of g/kg bw/d for the organophosphate/carbamate group (virtually 20 Regulation (EC) No 1272/2008 of the European Parliament and of the Council of 16 December 2008 on classification, labelling and packaging of substances and mixtures, amending and repealing Directives 67/548/EEC and 1999/45/EC, and amending Regulation (EC) No 1907/2006. fficial Journal L 353, December EFSA Journal 2012;10(07):

32 identical to the chronic TTC value of Kroes et al., 2004 for this group of compounds) and of g/kg bw/d for the remaining group of substances. The PPR Panel agreed that the current genotoxicity threshold of µg/kg bw/d (0.15µg/person/d) should be retained in the acute TTC scheme for substances with structural alerts for genotoxicity. Thus, the PPR Panel recommends acute exposure thresholds for pesticide metabolites of µg/kg bw/d for metabolites with a structural alert for genotoxicity, of 0.3 µg/kg bw/d for substances having structures suggesting neurotoxicity (AChE inhibition) and of 5.0 µg/kg bw/d for all other metabolites verall conclusions on TTC The PPR Panel concludes that the TTC approach is the most appropriate available tool in the evaluation of the toxicological relevance of pesticide metabolites associated with dietary exposure, for which there are few or no relevant toxicity data. This approach should not be used as alternative to conventional risk assessment for the evaluation of pesticide active substances (parent compounds) themselves occurring as residues in food. They should be assessed prior to authorisation on the basis of dossiers including toxicological tests (Regulation (EC) No 1107/2009). The PPR Panel noted that increasing numbers of studies address effects of chemical substances at low doses, with many of these studies referring to endocrine active substances or endocrine disruptors. According to the low-dose hypothesis, these substances may cause adverse effects at low doses but not necessarily at all higher doses. They do not therefore follow the classical (or monotonic ) doseresponse curve, showing a greater likelihood of an adverse effect at higher doses. Alternatively they may show a different kind of dose-response curve, e.g. a U-shaped curve with responses both at lowand high-dose levels but not in intermediate ranges. Such a dose-response curve is termed a nonmonotonic dose-response curve. Such findings challenge current concepts in chemical risk assessment including the TTC approach. n June 2012 (one week before the adoption of the current opinion) EFSA hosted Scientific Colloquium N 17 on low dose response in toxicology and risk assessment 21. Some key scientific questions and next steps in terms of methodological requirements, research gaps, appropriate testing strategies and methods, and the use of predictive tools were identified, but as yet no scientific consensus has been reached on the validity of the low-dose hypothesis. The PPR Panel concludes that for the time being, untested substances, other than steroids and several other categories of substances as concluded by the Scientific Committee 22, could be evaluated using the TTC approach. However, if there are data indicating that a substance may have endocrinemediated adverse effects, then the risk assessment should be based on the data, rather than the TTC approach. nce the EU-wide approach for defining and assessing endocrine disrupters is finalised it will be necessary to consider any impact it may have on the use of the TTC approach. The TTC approach seems to be widely held as scientifically valid in all regulatory areas where it has been considered, see 5.1 However the application of this approach is dependent on the quality and relevance of the underlying toxicity database, and a reliable estimation of the exposure to the chemical in the respective field of application. The PPR Panel considers that the TTC values for genotoxic ( µg/kg bw/day) and toxic compounds (0.3 µg/kg bw/day for P compounds, 1.5 µg/kg bw/day for Cramer class II and III and 30 µg/kg bw/day for Cramer class I) are sufficiently conservative for the evaluation of the toxicological relevance of metabolites, as a result of a validation process with groups of pesticides belonging to different chemical classes (CRD project). However TTC values 21 The outcome of the colloquium will be summarised in a report to be published in the autumn of The Scientific Committee concluded that the TTC approach should not be used for the following (categories of) substances: high potency carcinogens (i.e. aflatoxin-like, azoxy- or N-nitroso-compounds, benzidines, hydrazines), inorganic substances, metals and organometallics, proteins, steroids, substances with a high potential for bioaccumulation, nanomaterials, radioactive compounds, and mixtures of substances containing unknown chemical structures. EFSA Journal 2012;10(07):

33 based on the assumption of continuous lifetime exposure were considered overly conservative for acute exposure. Tentative TTC values for acute exposure were established by the PPR Panel by the analysis of the lowest 5 th percentiles of NAELs used to establish ARfD for the EFSA pesticide data set with values of 0.3 µg/kg bw/d for substances with a neurotoxicity alert and 5 µg/kg bw/d for substances allocated in Cramer class II and III. The TTC scheme proposed in the CRD s project work was considered by the PPR Panel as a suitable starting point to develop a decision tree for acute and chronic exposure to metabolites. Three critical steps identified in the TTC scheme were considered by the PPR Panel: a) the estimate of the level of the metabolite, b) the evaluation of genotoxicity alert that was addressed by the use of (Q)SAR approach; c) the potential neurotoxicity of metabolites derived from non-neurotoxic parents that was addressed by an ad hoc project exploring the use of computational methods. The PPR Panel considers that since there is, to date, no consensus on when a compound should be defined as an endocrine disruptor, risk managers have the following options with respect to applying the TTC approach, 1. not to use the approach until it is clear how to assess endocrine disruptor activity, 2. to use the TTC approach, but re-evaluate the applicability when there is consensus on how to assess endocrine disruptor activity. EFSA s SC recommended the latter option. In addition, the PPR Panel notes that at present there is no tool available for assessing the relevance of metabolites in a consistent way and that applying the approach described in the present opinion will improve the risk assessment even if it has to be adapted at a later date in the light of ongoing discussions on endocrine disruptors. 6. SAR/ (Q)SAR concept Computational methods, including (Q)SAR ((quantitative) structure activity relationships) and read across were considered by the PPR Panel as potential tools in assessing the toxicological relevance of pesticide metabolites in order to limit the need for toxicity testing in animals. SARs and (Q)SARs, collectively referred to as (Q)SARs, are theoretical models that are used to predict in a qualitative or quantitative manner the physicochemical, biological, toxicological properties and environmental fate of compounds from a knowledge of their chemical structure. The basic assumption for the application of (Q)SAR analysis in risk assessment is that the biological activity of a chemical depends on its intrinsic nature and in principle can be directly predicted from its molecular structure and inferred from the properties of similar compounds whose activities are known. More specifically, SAR is a qualitative relationship between a molecular structure or substructure and a specific biological activity, or the modulation of a biological activity imparted by another substructure. A substructure associated with the presence of a biological activity is called a structural alert. A (Q)SAR is a mathematical model (often a statistical correlation) relating one or more parameters derived from a chemical structure to a quantitative measure of a property or activity. (Q)SARs are quantitative models yielding continuous or categorical results. The parameters used in a (Q)SAR model are also called (molecular) descriptors. A molecular descriptor is a structural or physicochemical property of a molecule, or a part of a molecule, which specifies a particular characteristic of the molecule and is used as an independent variable in the (Q)SAR model Characterisation of chemical space The characterisation of chemical space is the first step in the evaluation of the adequacy of a (Q)SAR model as a predictive tool for a specific group of compounds. EFSA Journal 2012;10(07):

34 The chemical space of a dataset (or inventory of chemicals) is defined as the ranges of physicochemical properties and structural features covered by the chemicals in the dataset. The characterisation of the chemical space is relevant in the evaluation and application of computational models for the following reasons: because a model should be applied to chemicals within its applicability domain; outside of its applicability domain, a model is unlikely to yield reliable predictions; it is useful to compare the chemical space of the test set with that of the training set when the predictive performance of a model is assessed by challenging it with an independent (external) test set. Principal Component Analysis (PCA), a multivariate statistical method, is used to reduce complex multi-dimensional datasets to simpler lower dimensional datasets, minimising the loss of information Performance of (Q)SAR models The predictive performance of (Q)SAR models is generally assessed by the evaluation of the number of compounds correctly identified as positive or negative 23. Two parameters are considered to define the performance: the sensitivity and the specificity. The sensitivity expresses the percent of positive compounds correctly predicted and is calculated by the formula: Number of true positive (TP) compounds/(number of true positive compounds + Number of false negative (FN) compounds). The specificity expresses the percent of correctly identified negative compounds and is calculated by the formula: Number of true negative (TN) compounds/(number of true negative compounds + false positive compounds). The accuracy is defined as (TP+TN)/(TP+FN+TN+FP) Read-across A chemical category is a group of chemicals whose physicochemical and human health and/or ecotoxicological properties and/or environmental fate properties are likely to be similar or follow a similar pattern, usually as a result of structural similarity (ECD, 2007a). The grouping approach represents a move away from the traditional substance-by-substance evaluation to a more robust approach based on a family of related chemicals. Within a chemical category, data gaps may be filled by read-across, trend analysis and (Q)SARs (van Leeuwen et al., 2009). By its very nature, the grouping and read-across approach is an ad hoc, non-formalised approach based on a number of steps including expert choices. As with (Q)SARs (ECHA, 2010a) estimated properties obtained by the grouping and read-across approach need to be assessed in terms of their adequacy, and the justification needs to be clearly documented according to an accepted format (ECHA, 2010b). Critical issues in chemical category formation and read-across are the quality of the underlying experimental data for the analogues and definition of (chemical and/or biological) similarity (Jaworska and Nikolova-Jeliazkova, 2009) (Q)SAR approach in the dietary risk assessment of pesticide metabolites The PPR Panel addressed in more detail the potential use of computational methods in the evaluation of genotoxicity to complement the TTC approach Software tools for genotoxicity and carcinogenicity prediction Genotoxicity and carcinogenicity prediction is featured in a wide range of commercial and freely available software tools, as reviewed by Serafimova et al. (2010). The scientific literature relating to the in silico prediction of genotoxicity and carcinogenicity is substantial, with more than 100 papers dedicated to (Q)SARs. 23 This is only correct for categorical (Q)SARs having only two categories (positive or negative). Different parameters have to be used for (Q)SARs with multiple categories and (Q)SARs which give a quantitative result (e.g. a NAEL). EFSA Journal 2012;10(07):

35 A number of papers are devoted to the comparison of the performances of different models, including software models: many of them report the results of evaluation studies for prediction of carcinogenicity. The outcome of a series of external prediction exercises performed by various investigators with three models: MultiCase, TPKAT, and Derek were summarised (Benigni and Bossa, 2008). The common characteristic of these studies is that the chemicals to be predicted were different from those used in the training sets by the model developers, and were performed independently. It was found that the predictions for external chemicals vary considerably both in terms of overall accuracy and in terms of relative proportions of true and false positives. A factor which contributes to reduced model performance is the quality of the underlying mutagenicity data: inconsistent data interpretation or the lack of quality assurance may contribute to incorrect predictions made by in silico systems. When using computational models for regulatory purposes, the predictions of genotoxicity and carcinogenicity should not be based on the use of any single model alone, but on a weight of evidence approach including information from all available sources ((Q)SARs, read across, in vitro test methods). A number of studies in the literature (e.g. Contrera et al., 2007) support the usefulness of computational tools, applied in batteries that combine high sensitivity models (to minimise false negatives) with high specificity models (thereby minimising false positives) Current applications of (Q)SAR approach in predicting mutagenicity and carcinogenicity European Chemical Agency The REACH regulation 24 and associated guidance foresee the application of (Q)SARs in a number of ways (ECHA, 2008), mainly to: provide information for use in priority setting procedures; guide the design of an experimental test or testing strategy; improve the evaluation of existing test data; provide mechanistic information (which could be used, for example, to support the grouping of chemicals into categories); fill data gaps for classification and labelling and for risk assessment Danish Environmental Protection Agency To address the problem of classification under Directive 67/548/EEC and Regulation (EC) No 1272/2008 for the large number of existing substances in the EU not included in the list of harmonised classification and labelling of hazardous substances (Annex VI of Regulation (EC) No 1272/2008), the Danish EPA published an advisory list for self-classification of dangerous substances 25. The list of suggested hazard classifications was derived by using predictions from (Q)SAR models obtained or developed by the Danish EPA for the following endpoints: acute oral toxicity, skin sensitisation, mutagenicity, carcinogenicity. MultiCASE software was used for genotoxicity. Five different models predicting genotoxicity in vivo were applied for the screening by Danish EPA. The data for the training sets were obtained from papers available in the scientific literature. The 24 Regulation (EC) No 1907/2006 of the European Parliament and of the Councilof 18 December 2006 concerning the Registration, Evaluation, Authorisation and Restriction of Chemicals (REACH), establishing a European Chemicals Agency, amending Directive 1999/45/EC and repealing Council Regulation (EEC) No 793/93 and Commission Regulation (EC) No 1488/94 as well as Council Directive76/769/EEC and Commission Directives 91/155/EEC, 93/67/EEC, 93/105/EC and 2000/21/EC. fficial Journal L 13, May EFSA Journal 2012;10(07):

36 performance varied with the model applied: the sensitivity and specificity range from and from , respectively. For a substance regarded as a probable mutagen it needs to be positive in at least two models, accepting only predictions where no significant deactivating fragment was detected. If positive results from one or more genotoxicity tests were available, this would overrule negative predictions obtained with the models European Medicines Agency (EMA) The application of structure-activity relationships (SARs) is considered in EMA Guidelines for genotoxicity evaluation of impurities in pharmaceuticals. The absence of a structural alert based on a well-performed assessment (e.g. through the application of commonly used software tools including DEREK and MULTICASE 26 ) is sufficient to conclude that the impurity is of no concern with respect to genotoxicity. Compounds showing positive alerts not present in the active substance need to be tested with a bacterial gene mutation test. A negative bacterial gene mutation test overrules the structural alert. Structural alerts are also used in the context of the TTC approach. An approach proposed by Muller et al., (2006) in the application of computational models for genotoxicity prediction, uses a five class scheme to help decide whether an impurity possesses a high level of risk and should, therefore, be controlled at very low levels of daily intake: Class 1 Impurities known to be both genotoxic (mutagenic) and carcinogenic; Class 2 Impurities known to be genotoxic (mutagenic), but with unknown carcinogenic potential; Class 3 Alerting structure, unrelated to the structure of the active ingredient and of unknown genotoxic (mutagenic) potential; Class 4 Alerting structure, related to the active ingredient; Class 5 No alerting structure or sufficient evidence for absence of genotoxicity. It was demonstrated that DEREK for Windows can be successfully used as a first step for the identification of structural alerts for genotoxicity in the above scheme (Dobo et al., 2006). In a retrospective analysis of some 272 compounds, the implementation of this strategy gave an overall concordance of 92% for compounds in classes 1, 2, 4 and 5 with 67 (25%) of compounds falling into class 3, that would require further investigation Assessment of the equivalence of technical materials of substances regulated under Council directive 91/414/EEC The use of validated (Q)SAR models to predict toxic effects, including mutagenicity, is considered in the assessment of toxic hazard of impurities for the evaluation of equivalence of technical materials for substances regulated under Directive 91/414/EEC Applicability of (Q)SAR analysis to the evaluation of the toxicological relevance of metabolites of pesticide active substances for dietary risk assessment utsourced project by the Joint Research Centre (JRC) An outsourced activity was carried out by the Joint Research Centre (JRC) Ispra, to explore the applicability of computational methods in the evaluation of the toxicological relevance of metabolites of pesticide active substances (JRC, 2010). An extensive review on the available computational 26 MULTICASE was not part of the current evaluation (see Chapter 6) EFSA Journal 2012;10(07):

37 models with emphasis on (Q)SARs focusing on toxicological endpoints (acute and repeat-dose toxicity, including organ and system specific toxicities; genotoxicity and carcinogenicity; and reproductive toxicity; immunotoxicity, endocrine-related effects) and on Absorption, Distribution, Metabolism and Excretion (ADME) was produced (Mostrag-Szlichtyng and Worth, 2010). n the basis of the literature review, it was concluded that some available software tools (e.g. TPKAT and MCASE) are useful for predicting acute toxicity in categorical terms (e.g. in terms of Globally Harmonised System, GHS, classifications). The availability of (Q)SAR models for the prediction of chronic toxicity endpoints is very limited. Since a large number of potential targets and mechanisms are associated with repeat dose effects, it is unlikely that any single model or software tool will be capable of making reliable predictions for all chemicals of interest to dietary risk assessment due to limitations of the underlying database. The modeling of organ-specific and system-specific effects represents a developing field. nly a few (Q)SAR studies have focused on the effects of chemicals on the central nervous system, in some cases through the modelling of in vivo toxicity. Among the commonly used software tools, Derek for Windows v.11 includes the structural alerts: organophosphate (for direct and indirect anticholinesterase activity), N-methyl or N,N-dimethylcarbamate (for direct anticholinesterase activity) and gamma-diketones (for neurotoxicity). The availability of (Q)SARs for reproductive and toxicity (excluding models related to endocrine activity) is limited as a result of the diversity and biological complexity of the endpoints, and the scarcity of data suitable for modelling. Available models are potentially useful as a means of supporting hazard identification and priority setting, but not for use in risk assessment. A large number of (Q)SAR tools have been developed for ADME prediction, mainly for pharmaceutical purposes and for specific ADME properties (e.g. blood/brain barrier permeability, human intestinal absorption, placental permeability). However, their applicability in the dietary risk assessment of chemicals other than drugs is either poor or not established Framework for assessing the usefulness of (Q)SAR models In their report (JRC, 2010) the JRC introduces a conceptual framework for the use of (Q)SARs that is comprehensively described in the REACH guidance on Information Requirements and Chemical Safety Assessment (ECHA, 2008). According to the framework developed for REACH for using (Q)SAR models instead of experimental data these models have to be documented appropriately, must be scientifically valid, applicable to the chemical(s) of interest and relevant for the purpose they are used for in order to be considered adequate. In order to be considered as scientifically valid ECD (2007e) established the principle that any (Q)SAR model used needs to have a defined endpoint, an unambiguous algorithm, a defined applicability domain, appropriate measures of goodness-of-fit, robustness and predictivity and should, if possible, be associated with information on mechanistic interpretation. As a next step in assessing the adequacy of a model it needs to be verified if the chemical(s) of interest are within the applicability domain of the model. nly for chemicals that are within the applicability domain of a chosen model can reliable results be achieved. Therefore it needs to be assessed if the descriptor values of the chemical are within the predefined ranges and if it contains any structural fragments that are not known to the model, any predefined mode/mechanism of action and EFSA Journal 2012;10(07):

38 information on likelihood of transformation/metabolites and the characteristics of such transformation products. In order to demonstrate the adequacy of a (Q)SAR estimate generated by a valid and applicable model some additional considerations are needed, namely if the model endpoint is relevant for the regulatory purpose. Relevance of models predicting directly a regulatory endpoint is self-evident (e.g. (Q)SARs designed to predict LD 50 values). For (Q)SARs focusing on mechanistic endpoints, extrapolation from the modeled endpoint (e.g. nucleophilic reactivity towards DNA or proteins) to the regulatory endpoint (e.g. mutagenicity) needs to be made. verall, any prediction needs to be assessed within the context of the regulatory purpose and taking into account other relevant information applying a weight of evidence approach. The JRC has provided in their report a checklist with 10 questions that can give useful support in assessing the adequacy of (Q)SAR models. This checklist is inserted in Appendix F to this opinion Applicability of (Q)SAR analysis to the evaluation of genotoxicity of pesticide metabolites for dietary risk assessment: case studies In the JRC project, case studies were carried out on the potential applicability of (Q)SAR approaches in predicting genotoxicity with the aim of integrating (Q)SAR and the TTC approach Software tools applied The software tools were selected by JRC on practical grounds, taking into account their in-house availability, as well as budgetary and procurement constraints for the acquisition of new licenses. The software tools applied included: DEREK, a rule-based system combining toxicological knowledge and expert judgment; CAESAR, LAZAR, TPKAT, HazardExpert and ToxBoxes, based on statistical methodologies; Toxtree, a hybrid tool implementing both expert rules and statistical methodologies Compilation of datasets The ability to predict genotoxicity and carcinogenicity was based on the application of the various software tools to three datasets consisting of 185 pesticides, 1290 heterogeneous chemicals, and 113 heterogeneous classified mutagens Compilation of an internal pesticides dataset The chemical space of pesticides was represented by two datasets for which chemical structures were available: a) CRD pesticides dataset: initially consisting of 135 parent compounds from the CRD TTC study (see chapter 5) including 100 parent compounds used by the CRD to develop the TTC scheme, 15 to validate it and 20 metabolites. This was reduced to 128 after removal of structures that cannot be handled by computational tools (e.g. salts, organometal compounds); EFSA Journal 2012;10(07):

39 b) AGES pesticides dataset: initially consisting of 67 parent compounds from the AGES study (see chapter 4). This was reduced to 57 compounds after removal of compounds in common with the above-mentioned TTC dataset. The total number of case study structures, including CRD (128) and AGES compounds (57), was 185. Experimental data for carcinogenicity were available for 104 compounds (45 active, i.e. carcinogenic and 59 inactive, i.e. non-carcinogenic). Information on mutagenic activity was available for 181 substances, but only 11 of the compounds showed some evidence of genotoxicity in the Ames test, of which only 5 compounds (etridiazole, carbendazim, dichlorvos, thiobencarb, parathion-methyl) were associated with test data that might result in regulatory classification Compilation of external datasets a. Heterogeneous dataset: The DSSTox Carcinogenic Potency Database (CPDB) contains the results of cancer and Ames mutagenicity tests on 1547 chemicals (pharmaceuticals, natural compounds in the average diet, air pollutants, food additives and pesticide residues). From the initial database, the following compounds were excluded: inorganics (60), organometal compounds (44), compounds for which structures were not available, macromolecules, i.e. polymers, proteins, DNA, or other large biomolecular species (3) and formulations/mixtures (75). Since computational tools cannot handle certain structures (e.g. salts), these were also excluded, resulting in the removal of a further 36 substances, thereby leaving 1290 chemicals in the CPDB database. Carcinogenicity data were available for 1288 substances: 717 compounds were active (i.e. carcinogenic) and 571 were inactive (i.e. non-carcinogenic). Mutagenicity data, obtained using the Ames test, were available for 748 of the 1290 DSSTox molecules: 368 compounds were positive (i.e. mutagenic) and 380 negative (i.e. nonmutagenic). b. Dataset of classified mutagens: A series of 601 classified compounds was considered, comprising 594 substances extracted from the ex-ecb Claslab database and 7 substances added after comparing the ex-ecb database to Annex VI of CLP 27 (personal communication, J.J.A. Muller). From this, 113 substances that had been classified as mutagens (Muta. Category 2 R46 and Muta. Category 3 R68) during the EU harmonised classification process (the corresponding GHS classifications are Muta. 1B and Muta 2, respectively) were considered suitable for the analysis Characterization of chemical space by Principal Component Analysis (PCA) Principal Component Analysis (PCA) was applied to reduce complex multi-dimensional datasets to simpler lower dimensional datasets, while minimising the loss of information (variance in the data). Trends and patterns can be more easily identified by using the Principal Components (PCs), which are linear combinations of the original descriptors. The meaning of each PC can be derived from the loadings of the original descriptors on the PCs. For the purpose of this study, a range of easily interpretable descriptors (constitutional descriptors, functional group counts and molecular properties) were used. As a result, the chemical space was built from a combination of physicochemical properties and sub-structural features. PCA reduced the dimensions of 35 molecular descriptors to 5 representative PCs. The JRC study found that the chemical spaces of the pesticides (pesticides database including 821 compounds on the EU list of Plant Protection Products for which the structures were available) and the CPDB dataset were overlapping, thereby supporting the usefulness of CPDB when assessing the applicability of (Q)SARs to pesticides as well as other chemicals. The CRD dataset was broadly representative of the chemical space of the pesticide inventory, but lacking a number of structural 27 CLP is the Regulation on classification, labelling and packaging of substances and mixtures (EC 1272/2008). This Regulation aligns previous EU legislation on classification, labelling and packaging of chemicals to the GHS (Globally Harmonised System of Classification and Labelling of Chemicals). EFSA Journal 2012;10(07):

40 classes. Moreover, the use of a broader dataset increased the coverage of structural space, thereby providing a more extensive and robust analysis Performance of the models in predicting mutagenicity The performance of all of the models applied was assessed by the analysis of correct and wrong predictions. The number of compounds identified as true or false positive (TP, active and predicted as active and FP not active but predicted as active) and true and false negative (TN, not active and not predicted as active and FN, active but not predicted as active) was determined. Sensitivity was calculated as TP/(TP+FN) and specificity as TN/(TN+FP). The accuracy was defined as (TP+TN)/(TP+FN+TN+FP) Prediction results Prediction results for CRD-AGES dataset The results on genotoxicity prediction for the internal dataset including CRD and AGES pesticides are reported in Table 3. A 100% sensitivity was observed with ToxBoxes, although only 4/11 active compounds were correctly predicted and 7/11 were classified as equivocal. The lowest sensitivity was observed with Lazar: The sensitivity for the other applied tools ranged between 0.50 and The specificity ranged between 0.57 and The better performance of the model in identifying inactive compounds is related to the high percentage of non-genotoxic compounds included in the training and test datasets. Carcinogenicity prediction for this dataset is very poor: the range of sensitivity values is The specificity is higher, ranging between 0.53 and Table 3: Genotoxicity prediction for the CRD-AGES dataset Number of compounds: 185 Experimental values available: 181 Exp. active compounds: 11 Exp. inactive compounds: 170 SFTWARE STATISTICS* TP TN FP FN EQ ND SP SE CNC 1-SE 1-SP CAESAR Derek HazardExpert Lazar (Kazius/Bursi) Lazar (Toxbenchmark) TPKAT ToxBoxes Toxtree (Benigni- Bossa) TP true positives; TN true negatives; FP false positives; FN false negatives; EQ compounds predicted as equivocal; ND the number of compounds that were not handled by the software; SP specificity; SE sensitivity; CNC overall concordance; 1-SE false negative rate; 1-SP false positive rate Prediction results for DSSTox dataset The performance of the applied models for genotoxicity prediction was best for the external DSSTox Carcinogenic Potency Database (CPDB) dataset, including mutagenic and non-mutagenic compounds, classified on the basis of the results from the Ames test. The sensitivity values ranged between 0.66 and The specificity values ranged from 0.61 to ToxBoxes showed the highest sensitivity EFSA Journal 2012;10(07):

41 and specificity. This outcome is expected considering that the training dataset used to develop the applied (Q)SAR models is based on the same genetic endpoint, in vitro mutagenicity with Ames test Prediction Results for classified mutagen dataset The results for genotoxicity prediction for the external dataset of classified mutagens are reported in table 4. The highest sensitivity (0.87) was obtained with Toxtree, using the in vivo micronucleus rulebase, followed by HazardExpert (0.77). The lowest sensitivity (0.44) was obtained with ToxBoxes which is a model optimised to predict Ames mutagenicity. This result is expected because the dataset includes a large majority of compounds classified as genotoxic based on the in vivo micronucleus test. To improve the sensitivity of the applied models various pairwise software combinations were tested. If either tool in the combination gave a positive result then the overall prediction was considered positive. A reduction of the false negative rate was obtained, with the lowest value of 8 % for the combined use of Toxtree and Derek. Table 4: Genotoxicity prediction for the classified mutagen dataset Software (used alone) ND EQ TP SE FN 1-SE No TS Toxtree (genotoxic carcinogenicity) NA Toxtree (in vivo micronucleus) NA Toxtree (genotoxic carcinogenicity or in vivo micronucleus) NA TPKAT CAESAR HazardExpert Not known Lazar (Kazius/Bursi) * Lazar (Toxbenchmark) * Lazar (Kazius/Bursi or Toxbenchmark) * Derek (mutagenicity or chromosome damage) NA ToxBoxes Not known Software (used in combination) Toxtree or CAESAR Derek or CAESAR Derek or Lazar * Derek or TPKAT Toxtree or Lazar * Toxtree or Derek NA HazardExpert or CAESAR Test set of 113 classified mutagens; ND not determined; EQ compounds predicted as equivocal; TP true positives; SE sensitivity; FN false negatives; 1-SE false negative rate; No TS number of chemicals already in the training set of the model (where applicable); NA not applicable *: For Lazar it is not important whether a substance is in the dataset used to build the model, since an instance-based prediction is generated by a local model built from data that exclude the query chemical EFSA Journal 2012;10(07):

42 6.7. Conclusions of the contractor A conceptual framework for the evaluation of the adequacy of (Q)SAR models in the context of dietary risk assessment has been developed. A checklist was proposed and applied to select software models for prediction of genotoxicity of pesticides. See and Appendix F. An extensive review of the potential applicability of computational methods in the evaluation of the toxicological relevance of pesticide metabolites reveals that the usefulness of models for the prediction of chronic toxicity endpoints is very limited, while some available software tools are useful for predicting acute toxicity in categorical terms. A number of (Q)SARs tools for reproductive and toxicity and for endocrine distruptors have been developed as a means of supporting hazard identification and priority setting. A large number of (Q)SARs tools have been developed for ADME prediction, mainly for specific ADME properties (e.g blood/brain barrier permeability, human intestinal absorption). While it is difficult to give firm conclusions on the applicability of such tools, it is clear that many have been developed with pharmaceutical applications in mind, and as such might not be applicable to other types of chemicals (this would require further research investigation). n the other hand, a range of predictive methodologies have been explored and found promising, so there is merit in persuing their applicability in the field of food safety. The (Q)SAR case studies focussed on the applicability of several software tools for predicting genotoxicity of pesticide metabolites. The results of these studies, using the largest dataset available of active ingredients and metabolites, show a wide range of sensitivity from and specificity from The accuracy of the prediction is related to the training set data applied, as demonstrated by the high performance of ToxBoxes and Toxtree in detecting chemicals positive in the Ames test or in the in vivo micronucleus test, respectively. Several tools were good identifiers of Ames mutagenicity (typical sensitivities of ; typical false negative rates of ). Furthermore, some of these tools were good identifiers of classified mutagens (highest sensitivities of ; lowest false negative rates of ). Pairwise combinations of these tools could increase the overall sensitivity (to about 0.90) and reduce the false negative rate (to about 0.10). The software tools or combinations of them can be optimised with the aim of increasing the sensitivity, reducing the number of false negatives Conclusion by PPR Panel A checklist was proposed and applied to select software models for prediction of genotoxicity of pesticides. The panel considers the questions in the cheklist as valid ones, but did not use the tool itself. The PPR Panel concludes that the performance of the (Q)SAR tools applied individually, in the prediction of genotoxicity of the pesticide dataset, involving parent compounds and metabolites tested in the CRD and AGES case studies, is unsatisfactory (resulting in too many false positives and false negatives). The low sensitivity of the applied tools (between 0.45 and 0.64) could be attributed to the heterogeneity of the compounds in the dataset set. In addition the experimental data available to test the performance of the (Q)SAR tools are heterogeneous, including results on different genotoxic endpoints. Several tools were good identifiers of Ames mutagenicity with a sensitivity range of , some of them are also good identifiers of classified mutagens (sensitivities: ). The range of sensitivity and specificity values derived from the case study on the classified mutagen dataset applying (Q)SAR tools alone or in combination, is in the range of those described in the scientific literature. The results on the classified mutagen dataset confirms the usefulness of applying a battery of (Q)SAR tools to increase the level of predictivity. EFSA Journal 2012;10(07):

43 The usefulness of (Q)SAR tools in the prediction of endocrine disruptor activity was not investigated because of lack of a clear definition and availability of test results. verall, these outcomes, although not conclusive considering that one of the most commonly used software tool (MULTICASE) in genotoxicity prediction (EMA, 2006) was not explored in this exercise, do not support a proposal for the application of a (Q)SAR approach alone to predict the potential genotoxicity of unknown pesticide metabolites. This conclusion is in agreement with the considerations reported in the EFSA SC opinion on genotoxicity testing strategies (EFSA, 2011b). However, the PPR Panel recommends that the application of integrated approaches including combined (Q)SAR models and read-across is explored in future studies. The use of read-across implies the availability of a robust database comprising the main genotoxic endpoints. Further research is needed to develop batteries, including (Q)SAR models for each critical genotoxic endpoint, with the aim of increasing the sensitivity, and reducing the number of false negatives. Two important challenges faced by (Q)SAR models for genotoxicity prediction of pesticide metabolites are the diversity of compound structural space including the differences between stereoisomers and the multiplicity of structural alerts that can produce the same effect. The development of mechanistic SARs and the possibility of expanding the applicability domain could increase confidence in the predictions made by in silico models allowing improvements in the future use of (Q)SAR model combinations in the prediction of genotoxicity. The outcome of the (Q)SAR project allows the PPR Panel to propose the application of computational methods, involving separate or sequential use of (Q)SAR and read-across, as a complement to the TTC approach in the assessment scheme for pesticide metabolite exposure. If the analysis predicts genotoxic activity, then the metabolite is by default considered as genotoxic and the choice of further testing to prove otherwise rests with the applicant. If the analysis is negative, further testing is still required because of the probability of false negatives. The use only of computational tools, (Q)SAR and read across, should not be employed in the evaluation of pesticide active substances themselves (parent compounds) occurring as residues in food. They should be assessed prior to authorisation on the basis of the results of a battery of tests using a stepwise approach. 7. Applicability of (Q)SAR analysis in the evaluation of and neurotoxicity effects of pesticide metabolites: outcomes of outsourced activity Within the EU peer review of active substances, the need to establish an Acute Reference Dose (ARfD) on the basis of adverse effects exerted early in repeat dose toxicity studies, is most commonly triggered by either or, albeit to a lesser extent, neurotoxic effects. A similar conclusion was reached by Solecki et al., (2010) in a recent review of on ARfD setting within the EU. The outcome of the (Q)SAR project carried out in preparation of the opinion (see Chapter 6) suggested that computational tools could be used to explore and neurotoxic alerts. EFSA therefore commissioned a further project at the JRC in which computational tools were evaluated for their suitability in excluding and neurotoxic effects of pesticides with the aim of possibly including them for refinement of a draft assessment scheme for metabolites. A stepwise strategy was followed in which (Q)SAR tools were used in an initial step for the identification of potentially neurotoxic active chemicals and a subsequent step, based on grouping and read-across was applied to discriminate between true and false negatives generated by the (Q)SAR analysis. EFSA Journal 2012;10(07):

44 7.1. Predictive performance of (Q)SAR/read-across strategy for neurotoxicity A dataset for neurotoxicity, including 40 positive and 21 negative substances was provided by EFSA. Twenty-one substances among the positives belonged to different chemical classes, carbamates, neonicotinoids and pyrethroids, which are expected to show neurotoxic effects. rganophosphates were not included in the dataset because the neurotoxic mechanism of action of these compounds is also well known in humans. In addition, such compounds are readily identified from their structure alone. The available software tools for predicting neurotoxicity were considered. The large majority are commercial and some are related to prediction of blood-brain (BB) barrier penetration and are not directly relevant to the current project. Five (Q)SAR models were applied as a first step: Derek Nexus, HazardExpert, Pass, ADME Predictor (probability of blood-brain barrier permeability predicted as low or high), and Accord (quantitative linear regression model for predicting the blood-brain barrier penetration as log BB). The performance, in terms of negative predictivity, of the individual models was low: the best performing tool was Derek with a specificity of 100 %, a negative predictivity of 43% and a false negative rate of 74%. The use of two-model combinations increased the negative predictivity to 48% (for the combination Derek and Pass), but this was also associated with an increased false negative rate (84%). The read-across approach was not used for the neurotoxicity prediction due to the lack of a suitable reference database Predictive performance of (Q)SAR/read-across strategy for toxicity Three sets of data were considered for the predictivity of toxicity: A) A dataset of pesticides provided by EFSA including: 37 pesticides positive for effects, identified by considering the substances for which the ARfD was based on toxicity and selecting early onset specific malformations in rat and/or rabbit at maternally non-toxic doses considered for the establishment of an ARfD; 39 pesticides negative for effects, with no adverse effects observed in valid development tests with rat and/or rabbit at doses up to those associated with maternal toxicity; B) An extended dataset of 135 substances, which comprised the EFSA dataset of pesticides and an additional group of compounds classified for toxicity, provided by RIVM; C) A dataset derived from the US EPA s ToxRefDB. The performance of seven (Q)SAR models suitable for toxicity prediction was evaluated, in terms of negative predictivity (see Glossary) in order to exclude the metabolites from acute exposure assessment (Table 5). The results based on the EFSA dataset suggest that Derek, TPKAT and PASS are the best stand alone tools in terms of their negative predictivity, although these models cannot be considered suitable for use on their own, due to their low negative predictivity ranging from 49-55%. Analysis of a larger dataset, US EPA s ToxRefDB (in this case the substances were considered positive for any adverse effect when the Low Effect Level (LEL) was lower than the maternal LEL), which included different categories of substances, showed that negative predictivity (87%) and false negative rate (37%) were best with Leadscope. The evaluation of the predictive performance of batteries of toxicity models showed that the best results were obtained with HazardExpert combined with PASS, with a specificity of 100%, a negative predictivity of 41% and a false negative rate of 74%. EFSA Journal 2012;10(07):

45 Table 5: Predictive performance of toxicity models used alone against EFSA test set (taken from JRC, 2011) Derek Caesar TPKAT Leadscope Hazard Expert PASS (embryotoxicity) PASS (teratogen icity) A * B A B A B A B A B A B A B % of chemicals sensitivity specificity concordance negative predictivity positive predictivity false negative rate false positive rate No of chemicals (A 76, B 135 in total) TP TN FP FN ND *A = EFSA dataset of 76 pesticides for which ARfD was based on effects; B = Expanded EFSA dataset of 135 chemicals. As a second step of the project, the suitability of the read-across approach was explored. The consistency of this approach depends on the selection of appropriate analogues and on the availability of reliable experimental data. ECD (Q)SAR toolbox was used for grouping the compounds and the EPA s ToxRefDB database was selected to perform the read-across exercise for toxicity. The aim of this exercise was to refine the predictions resulting from the application of (Q)SAR. Chemicals can be defined as active, inactive or inconclusive, as a result of a number of expert choices in the read-across procedure. The predictive performance of this tool cannot be evaluated because the outcomes could be different on the basis of different choices, but it could be considered in a step-wise approach combined with the use of (Q)SAR tools. Table 6 (Table 9.1 of the JRCreport, 2011) shows the possible outcome of a proposed stepwise strategy carried out with the EFSA extended dataset (135 substances) and involving the application of the PASS model for teratogenicity, then grouping and read-across using the ECD Toolbox and US EPA s ToxRefDB database. The overall outcome shows that read-across increases the positive predictivity of (Q)SAR analysis (to 90%), allowing better discrimination of true and false negatives generated by the use of (Q)SAR. Table 6: Possible outcome of applying the stepwise assessment strategy (taken from EC, 2011) Step Entering Predicted positive Predicted negative Not predicted Filtered out Proceeding to next step 1. Existing data P,39N (adequate data) 92 72P,20N EFSA Journal 2012;10(07):

46 2. (Q)SAR model (PASS teratogenicity) 3. Read-across (ECD Toolbox) (no data) Totals 68 62TP,6FP 9 8TN, 1FN Conclusions of the contractor No individual (Q)SAR model or combination of models appears to be adequate to predict the neurotoxic potential of pesticide metabolites, based on the outcome of the exercise carried out with a limited dataset comprising 40 positive and 21 negative pesticides. The application of a stepwise approach including (Q)SAR models and read-across for the prediction of the toxicity of pesticide metabolites appears to be promising. The key step for future development of this strategy is the establishment of a searchable structural database including high quality toxicological data on pesticide parent compounds. In addition, considering that toxicity is a complex process involving short term and long term effects associated with acute and/or chronic exposure, classification of the events could help in the identification of acutely toxic substances. A stepwise assessment scheme based on the combined use of (Q)SAR and read-across was proposed. The general stepwise assessment scheme based on the use of existing data and non-testing methods in the report from the contractor (see reference list). A first step involves the use of (Q)SAR models, alone or in combination, for identifying toxicants. A second step includes a further evaluation of the compounds predicted as negatives by (Q)SAR using a read-across approach Conclusions by the PPR Panel The PPR Panel concludes that the predictivity for neurotoxicity of the (Q)SAR models, tested alone or in combination, is currently inadequate to be applied in the evaluation of the toxicological relevance of metabolites. A similar conclusion was reached with the application of DEREK in the TTC case study performed by CRD, where a more reliable prediction of neurotoxicity was obtained using the known mechanism of action of pesticide active substances. It is not possible to use a read-across approach to predict neurotoxic effects at the present time, due to the lack of reference databases for this effect. (Q)SAR tools alone are not sufficiently reliable to predict effects, due to their low negative predictivity. The read-across exercise performed by the contractor as part of a stepwise approach, using the US EPA s Toxicity Reference Database (ToxRefDB), which includes data on toxicity for more than 300 pesticides, resulted in an improvement in the identification of potential toxicants and non- toxicants. The PPR Panel considers that a combined approach including (Q)SAR and read across, as proposed by the contractor (see chapter 11), could be a preliminary step before the application of the TTC scheme, in order to evaluate if an acute exposure assessment is required. No clear criteria could be derived for the application of the proposed scheme, because the read-across approach is not an automatic procedure; unlike the (Q)SAR tools it is very dependent on the selection of the reference database and a number of expert choices involved. Research is needed to further develop the use of (Q)SAR tools, by classifying the different endpoints associated with toxicity. In addition the development of an appropriate database on pesticides, would allow improvement in the use of the read-across approach. EFSA Journal 2012;10(07):

47 8. Potential exposure to pesticide metabolites in the human diet Toxicological relevance of pesticide metabolites 8.1. Introduction In order to evaluate the relevance of pesticide metabolites it is necessary to have conservative exposure assessments, which take into account high exposure scenarios. The potential for exposure to metabolites in food and feed may vary depending on active substance and residue specific factors as well as the exposure scenarios being considered. In this opinion chronic and acute exposure scenarios considering metabolic profiles in different crops and various intended uses of the pesticide are presented. Residue levels of pesticides and metabolites in plant and livestock depend on several factors related to Good Agricultural Practice (GAP) i.e. mode and time of application, applied dose, number of applications and the pre-harvest interval (PHI), but also on environmental conditions and nature of the crop or livestock. The pesticide may be converted into metabolites to different degrees quantitatively and also qualitatively, due to differences in metabolic pathways across different species. When estimating the exposure to the metabolites with limited data available there is a need to extrapolate between different crops and potentially between different crop metabolism groups and to consider the extent of uses that should apply. The levels of metabolites in various crops depend on many factors, especially pre-harvest interval, type of crop and part of plant considered, and therefore the relative amounts of metabolite(s) in relation to parent seem to follow complex rather than easily predictable patterns. This is also reflected in the practice of setting the conversion factors, for applying to the level of residue in the residue definition for monitoring to one for use in risk assessment, as presented in this chapter (chapter 8.5) and demonstrated in Appendix D. The estimation of exposure to metabolites has been tested by the Panel, building on the exposure work presented in the outsourced project (CRD, 2010), by way of case studies, encompassing scenarios for primary crop, rotational crop and livestock on six pesticides, namely azoxystobin, bitertanol, boscalid, dimethoate, napropamide, and prohexadione calcium. The case studies are presented in Appendix E. In Figure 1 an Exposure tree describes the different steps to consider when estimating exposure to the metabolites of a pesticide. The different steps in the Exposure tree are explained in detail in this chapter and illustrated in the case studies, see Appendix E. EFSA Journal 2012;10(07):

48 Residue Scenarios Ratio of metabolite to parent C H R N I C Use STMR Total Dietary Intake A C U T E Use HR Acute Intake for each Commodity To all crops /crop groups Critical consumer Intended use Ratio of metabolite to parent Pesticide uses Crop(s) in which metabolites were found only Representative use Test TTC Chronic TTC levels Acute* TTC levels * Required unless acute concerns can be discounted for a metabolite Figure 1: Assessment scheme for chronic exposure Residue scenario: Metabolites may be qualitatively and quantitatively different in various residue scenarios i.e. in primary crops, rotational crops or in animal products, see case study Appendix E Ratio of metabolites: nly limited data on levels of metabolites are usually available. Therefore consider the level of metabolite to a level of parent as a ratio for use in estimating chronic and/or acute exposure, see chapter 8.3 Chronic and acute exposure estimation: For chronic intake calculations of metabolites use STMR of the parent and for acute use HR of the parent in conjunction with the metabolite ratio, or where necessary alternative methodology discussed, see chapter 8.3 Critical consumer: Calculate the exposure for various consumer groups, including young children and other vulnerable groups, and deduce the highest result across all of the consumer groups, representing the critical consumer. Pesticide uses: Ratio of metabolite to parent may differ with PHI, crops and crop groups. The extent of pesticide use on different crops may also differ. These factors influence intake estimate considerably, see case study Appendix E Test TTC: Compare the intake estimates with relevant TTC, see Appendix E and G EFSA Journal 2012;10(07):